AI Agents & Automations for E-commerce [FULL COURSE 2026]

Take a look at this chart. Major AI model releases in 2025 and their capability. AI is getting smarter faster than most people realize. Most sellers think AI is just another tool in their kit. But mastering AI is like a meta skill. In other words, if you get good at using AI, everything else in your business improves and accelerates. Your marketing, your operations, your product development, all of it. I've worked with Dozens of seven and eight figure sellers over the last year and one thing is clear, most sellers heavily underestimate what current AI models can do.

So, I put together this course that will get you ahead of 90% of sellers and most importantly prepare you for what's coming in 2026. You will learn how to master prompt engineering, which is the foundation of every AI agent and automation. You will learn how to connect AI to all of your business tools Using MCPs and how to build powerful AI automations with N8N. This is a comprehensive course to teach you everything you need in just a few hours. So, let's dive in. Before we build AI agents and automations, we need to master the foundation,

prompt engineering. This is the skill that makes everything else work. If you can't get good outputs from your prompts, your agents won't perform and your automations will break. We're Covering the prompt engineering framework, chain of thought reasoning, extend the thinking mode, and how to structure your instructions so your AI actually does what you want. By the end of this section, you'll know how to prompt like a pro, and everything you build after this will work because of it. Let's start. Did you know that Ced GPT has its own official prompt optimizer? Let me show you

how it works. So, if we hop in here into the optimizer And leave a prompt we would usually use, let's say, I want to launch a product on Amazon. Here's the keyword list. help me generate the listing. And then we click optimize. It generates something that looks like this. It's completely different. There are suddenly some roles and objectives, task, checklist, instructions, even some context. So suddenly it turns it into a prompt that would generate much better output. But not only that, it also allows us to Review the changes and the reasoning behind these changes. So

we can read why it's making these changes and learn ourselves how to prompt better in the future. So why should we even care about getting better at prompted? Well, a lot of people including me when CH GPT5 was released, the latest model were complaining that the model is not as good as the previous one, the GPT40. Well, and they were right because with the old prompting techniques, CE GPT5 Actually performs worse than the previous model. But it doesn't necessarily means that GPT5 is worse. It's actually quite the opposite. It's built with AI agents in mind,

which means that it's much much better at following specific instructions. Well, and when we don't provide the instructions and the context, we are not going to get a good output. So that's why the prompt optimizer helps us optimize our prompts, which is very Useful. But more importantly, it also teaches us why it's making these changes. So as you can see it's actually optimizing it for this specific model. In this case the reasoning behind the change it's adding conceptual checklist as a planning step supporting GPT5's plan first rule. So it understand how the new model works

and generates the prompts in a way to generate the best output possible from this particular model. And as you probably know, all Large language models like CHGPT or CLA actually predict words based on patterns and context. And this is very important because we need to remember that the clearer the context, the better the prediction. In other words, a simple rule is that your AI outputs is only as good as your prompt. In other words, garbage in, garbage out. And I always like to remind myself that this is actually like hiring and onboarding a new employee.

Compare the performance Giving someone three words versus giving them a full step-by-step SOP. It's exactly that. So the first step before we dive into the advanced things I'm going to teach you today is actually understand prompt engineering on a high level. So when structuring your problems essentially how you interact with uh with a GPT I like to think about this as building blocks and this simple formula will kind of give you the framework for writing prompts as an expert. So first Of all we should always specify the role who should check GPT act as then

we specify the task what is the outcome what is our objective and what are the steps as you can see this is kind of similar as writing an SOP for uh for an employee then we give it some context you know what do we already know what should be considered are there any constraints and lastly the format like what should the output look like is it a table is a bullet list is an SOP is it a JSON you know whatever that is we need to specify that and for that actually we have uh from

open AAI their cookbook and if we go here there is a full prompting guide you don't need to read this I actually pinpoint the most important points for you to remember to kind of get the 8020 but if you are interested there are more and more specifics to understand how the model works and more importantly there is actually the prompt optimizer and I will include the link Down in the description so you can use it yourself But the most important step when prompting and interacting with CHGPT5 is triggering the chain of thought reasoning. It's essentially

a sequential thinking mode that will unlock the deeper logic, fewer mistakes, and more structured output. So, let me show you how to trigger the chain of thought reasoning. The first one is a router nudge. Let me explain what it actually means. Because every time you Send a prompt to chat GPT, imagine that there is an invisible router that decides how much effort the AI is going to spend on your prompt to give you the output. Let's say if you ask for a simple recipe, it will show you low effort to do that because it's very

simple, right? But if it's a little bit more complex, it might use medium effort. And if it's very complex task that it needs to think deeply about, it will use high effort. And that will Trigger the chain of thought reasoning. And we can use these phrases in our in our prompts to actually trigger these sequential thinking mode. And it can be truly as simple as this as including these phrases at the end of your prompt like plan and outline first, think step by step, or even just think deeply about this. And you can test this

yourself to see the difference. If you let's say I want to explan expand from US to Europe. I need your help to create my action Plan. Before creating the plan, show me your thought process and outline how you would do that. And this pretty much guarantees that the AI is going to spend high effort in order to do that because you need to keep in mind that on their end it's a business and what they want to optimize their cost. So in most cases they will try to send it to the loweffort model right they

don't want to waste that because you are paying subscription you are not paying per Action so they will in most cases send it to the low model because most people are simply not going to specify and do these things to actually trigger the high level of reasoning. So the router notch is the first one but the second one which you can activate by clicking a button is actually extended thinking. when we can activate the thinking mode deliberately. So it actually activates the internal reasoning before it crafts its replies. And this is extremely Useful for complex

and strategic tasks like most of the things that you will use in business. So you will simply do it like this. Most people interact with CHGPT like this that they leave the model on auto and this is essentially leaving it up to the router to decide how long to think and as we said it's a business so in most cases if they can they will send it to the low effort but what we are going to do we are going to switch to thinking and not only that we Are going to switch to extended thinking

right this might take it a minute before you get your answer but really we don't care about the speed as much in most cases. We care about the accuracy. So, let me show you what's actually the difference if we compare the exact same prompt and we ask uh CH GPT, how should an Amazon seller price their product to stay competitive but still profitable? And this is with the quick answer with the low effort, you know. So, you get Some start with cost clarity, analyze competitor pricing, set your initial price. like this is good advice but

it's very surface level right this is kind of like yeah here you go you know that that's kind of what you ask for right but no you are a business owner you want more detail you want something more tangible more strategic so if we switch to the extended mode extended thinking look at the difference exact same prompt he thought for 25 seconds right I wasted 30 seconds of my life but I got this level of detail suddenly we are getting build unit economics first right we are getting the formulas how to calculate everything position against

the market you know choose your lane are value parity or premium product test price deliverables it also accounts for price elasticity right it's very advanced and it's kind of the same answer right he's answering the same question but in a very very different way so if there are Only a few things you would remember to get better outputs almost instantly just use these phrases you know plan and outline first and use the extended thinking mode. That will unlock way way better output. And I want to also stress out that it may seem that longer prompts

are better prompts, but it does not necess that isn't necessarily the case, right? Because as we talked about this, we care about the context, right? How much information we provide and how Specific we are with our instructions. But it also has a downside, right? We don't want to dilute our requests, right? If you have these two, three pages of prompts, you need to understand that there is such a thing as a context window in your chat and it kind of dilutes the information, right? So rather than thinking in longer prompts, think about being as specific

as possible. Now, we understand the basics of prompt engineering, kind of the 8020. But I know what you're probably thinking, right? You don't have time to do this. Every time you interact with CH GPT to write prompts in this structure to go into your prompt optimizer to create the perfect prompt, you know, you don't have time to do that. And I completely understand. That's why we will introduce this concept of context stacking because in most cases the context doesn't have to change as much. That's why we will introduce these Prompts. The first one is a

master prompt. This one doesn't really change because it is us introducing ourselves to the AI. So the AI knows who it's interacting with. This in other words is our identity. The master prompt. This defines who we are, our perspective, our goals and our preferences for communication style. Right? Let's say I'm a founder of an Amazon private label brand. We're doing 5 million a year. My goal is to get it to eight figures and Exit. You know, these are the products I sell. these are the marketplaces, this is my team and so on. So the AI

has the context, we do it once and then we get the benefit of this, right? And when we are creating something we will reuse again and again, well, we can spend the extra time to optimize the prompt and make sure that it is as accurate and as possible. [snorts] Now if we go a little bit deeper, we will have the system prompt. This is the role for the AI. In This prompt, we define who the AI is, the function, the responsibilities, and the boundaries of the agent. Right? Because before we assign some task, we need

to do two things. Who am I? What are my goals? Right? But also, who are you as the AI, right? Maybe you are my uh content strategist. Maybe you are my tax advisor, right? It is important to specify and clarify the role of the AI for the interaction. And lastly, we have the user prompt, right? So that is the Task and that defines what the AI should do, the goal, the steps and the examples. So as you can see, most of this five-part formula, the role, the the context, the constraints and so on is already

predefined in our reusable prompts, right? And of course, this structure, I want to show you how to build this. And I call this your AI workspace, right? And we are going to cover this now. how to build this out for you and your Amazon business and Ideally for every single person on your team because the leverage is just not there for you. The real leverage comes from everyone on the team using this and building out this foundation, right? So, let's take a look at it. When we want to build our AI workspace, how does this

actually work? Because we're talking uh master prompt, system prompt, user prompt, where where does it go? Like, should I always write everything here? Well, no, because chat GPT actually has Uh multiple functions that we can leverage that are basically built for this what we just discussed, right? So, the first one is personalization. Maybe some of you already tried this. When you go into settings, you will notice that there is a section for personalization, right? So, there is some style and tone how the AI should communicate, right? You can use the default or robot and so

on. I actually uh for business purposes I prefer the robot. And then there are Some custom instructions here. I would not use this as a space for master prompt. I would actually use this to really clarify the communication style I prefer. So let me explain. If we go here, you're going to build this together. In the personalization style, I would include some of these sentences that actually explains what we like. Because as we said by default the chet GPT is actually designed to kind of agree with you to make you feel good to Predict what

you want to hear which is the exact opposite of what we want in business right we want it to be honest not reassuring we want it to be direct unfiltered no fluff no validation right so we first need to set this up for most of our interactions right so for example I prepared some of these sentences here for you can test it out for yourself but I for example use just these prioritize accuracy over agreeability, be honest, not reassuring, be direct and Unfiltered, no fluff, no validation. And this helps me to get much better outputs

across all of my conversations. But you need to keep in mind that this personalization applies to all of your interactions and sometimes it might not be desirable, right? Maybe we are creating marketing content. We don't want this, right? We want a different style of language. So ideally this level of personalization I would include in the master prompt or the system prompt Which we are going to write later on right. So you can leave these empty. Uh then we have the nickname, your occupation and so on. So this is not as important and not going to

change how we interact with the AI. But now finally we are getting to the important part which is the master prompt. Right? This is giving chat GPT a permanent understanding of who you are and how you operate, right? Ideally in the context of business if you use it for business, Right? So your uh this master prompt should include things like this. Your role in the business, you are the founder, but if your employees should do this as well, which I highly recommend, they can be I'm a supply manager for this brand doing 5 million a

year. These are my responsibilities. These are my KPIs and so on. Right? So there will be some business overview. What is the business about? The niche, the revenue level, team size, goals and priorities, What success looks like, and potentially tools and data sources you work with. You know, it's like I use with Helium 10, you know, I work with Google Sheets to run manage my inventory and so on. So these specifics about you right now might be quite intimidating to actually build this, right? Because it's it's a lot of things and I don't know how

the master should look like. Well, that's why we have AI. it can kind of help us to create this prompt for future Conversations. So, let me show you something, right? If we have this prompt, just a basic basic uh prompt I would write when I want to create my master prompt. I'm a founder of an Amazon brand selling Italianmade shirts for men. I need your help writing my master prompt. It should include details about my business, my goals, and my responsibilities so we can reuse it in future conversations. ask me questions to clarify everything you

need to write It for me. And since this is something important I will reuse, I kind of want to use it as a meta prompt, right? So I want to take this prompt I wrote and first before I start my interaction, I want to go into the prompt optimizer and optimize this prompt, right? Just to make sure just to try this out a bit. It will take a few uh few seconds to load. And in the meantime uh what we will do once we have this prompt just to remember we will save it As a

document right because we will be working with this master prompt in our future conversations so we need to remember that but I also like to create uh essentially a meta project that where I will store my prompt generators because we will be generating a master prompt we will be generating systems prompt and so on. So what I like to do is to hop in uh CH GPT go into new project it's very simple uh just choose some nice icon and here I will go prompt Generator right here I create the prompt generator where I will

be generating all of my uh all of my master prompts and the system prompts and here when I create the project I can go into add instructions and don't worry we will get into a little bit more detail later on, but for now I want to include it here. And include a very simple um instructions. You are an expert prompt engineer specializing in creating prompts for AI language models Particular JP5 thinking model. Output final prompt in markdown format. Right? So this is what we will do. Um and also it helps us store all of the

conversations for future reference. Right? If we want to change our master prompt, we can find the chat in this project. So we can make the necessary changes. And now we can see um that he made the change to our prompt where we were requesting to help us with the Master prompt. And as you can see, he made some changes, right? Um as you can see, he made it a added a little bit more detail, right? So I need your help to craft comprehensive master problem that will serve as a reasonable alterative reference in future conversation

begin with a concise checklist and so on. Um, so now I can take this. I like it, right? And I would put this um into my prompt generator. So I can start generating my um my master Prompt. And here I would recommend to spend some time, you know, spend hour or two to actually make sure that it includes all of the details. Kind of refine it a few times, which of course I'm not going to do live. But what I'm going to do is that I'm going to pretend like I answer all of his questions,

right? Sometimes when you click as as he's thinking, you will see the thought process behind it, which is also interesting sometimes to actually see How the how the AI works, right? So he is thinking to create master prompt, final prompt creation, right? So you can see the thought process and I think this is helpful because now it's clear that we triggered the chain of thought reasoning, right? So we will wait a bit uh for him to come up with the questions. we will be answering a bunch of questions. So once we answer them, he will

be able to generate uh our final master prompt and once we do that we Will make uh we will take the next step which is creating the system prompts for chat GPT. So let's wait for a bit and now we can see what he came up with right so here's the plan and the deliverable I will build a checklist capture brand snapshot set objectives define KPIs and target guardlines map constraints and so on. So now he's start asking us a bunch of questions right target customer uh how we actually what is the price of The

brand and so on our markets and channels the brand positioning one line top three differentiators competitor you benchmark margin model right so I can update some data so he gets as much context as possible and then he will create um the master prompt here you can kind of see what the master prompt will look like once we give him all of the information and as you can also notice that he he includes it as a code and this is an important thing because here Output final prompt in markdown format the markdown format you may notice

that uh he actually did it here as well that is using just the hashtags which essentially means that he's creating headlines right this is how the works so it helps us to understand what is a headline what is a text and so on so when we request a markdown format he already kind of creates um these uh these separators, you know, what is the role and so on. And you also may notice That he's kind of following what we discussed before. What is the optimal structure of a prompt, you know, the role u act as

you know and so on. So here we will just say uh for the purpose of this video that assume we answered all of the all of your questions and now give me the final master prompt. So now we simply populated the master prompt with dummy data you know um I told him to assume that I answered all of his questions and here can be role hey um MKL founder brand owner and Amazon seller you are my brand co-pilot and so on so uh this would be the master prompt of course I would spend some time

refining this because this is one of the key assets that I would be reusing in most of my uh business interactions with the with the AI in some cases I would also clean it up a bit maybe add a little bit more detail but let's let's assume that this is it and I would create a simple text file right so I can Save it for future reference or you can save it in your notion whatever and here I will copy and paste and save it for future reference now I have my master prompt so we

can continue right we can continue with the system prompt and now the system prompt as we said defines how chat GPT should think, behave and format the output. Right now we are focusing on the AI and in this case it's not always the same because it depends on our task and what we want from it. So what chat GPT offers, it offers projects, it offers custom GPTs uh and so on. There there are these functions that we can work with. Now we created the project but we also have the ability to create custom GPTs, right?

So where do we begin and how do we create this? So I have a simple rule for this so you know when to use what because they are slightly different and I will cover all of the differences. So as as I already showed you we have these Projects right I use projects for thinking and building. This is where you plan iterate and upload you know your documents. It is an ongoing conversation that can evolve over time and it literally serves for projects, right? So let's say you have a brand revamp project, right? You decided to

change the target audience, the visual style of your brand. So this would be perfect for a project because it's evolving. You're building assets, you Know, your target customers, you are revamping your listings. So this is a great use case for a project. So we would do uh something like this. The first step would be creating the system uh the system prompt for this project. So we will go into our uh into our prompt generator and we here we have I'm working on a brand reven for my Amazon business and I want to create a system

prompt for it. Help me write it. Ask me any questions you need and then create a Simple system I can use for all future brand strategy and creative work in this. Uh so I can say this and then uh my master prompt details about my business is attached right I can already uh go here and upload the master prompt so he knows what he's working with so I went into files add file and uploaded my master prompt right it's attached in the project files for reference um and now let's see right now we are creating

the system prompts which are Essentially the instructions he has in the context window for every interaction within that project. Right? I know it may sound complicated but it actually is fairly straightforward. So now he's reading the document. Now he's uh asking us questions how to create the system prompt any subbrands lines or seasonality themes hard limits on price brand and margin floors per as visual style rules beyond refined confident minimalist you know describes to avoid. So now he's um asking us questions again. We would tell him that for this case he actually created uh a

draft of the system prompt for the brand revamp uh project. Uh so I will again tell him assume I answered everything and create uh a system prompt with dummy data. So here we have the dummy system prompt for our brand revamp uh project. So I'm going uh to copy all of this and we can create the new project. So as you saw before I like to create a new project. It will be brand revamp project. Uh we will choose some visual right here. And also you can choose whether the memories should be saved in your

you know uh outside of the chats with with your whole workspace or project only. In most cases, I like to choose project only, but it really is a matter of preference. And now we create the project. And once the project is created, we will be able to go here, add instructions and copy paste this long long uh system prompt, Right? What I also then do is just for reference for this particular case, it makes sense to again upload our master prompt. So I would go here and upload the master prompt. I can potentially fit it

into the instructions. But the main difference here is that the files that you upload here are accessed on demand. They are not uploaded directly into the context window. In other words, the AI doesn't know what's in these files unless you ask him directly or Indirectly to access them. Right? So in most cases, the master prompt, he will access it. But let's say there is there will be a brand voice document specifying how you communicate in uh in your brand you know your your brand assets and in case it doesn't have to get access to it

it's not going to open it and in the instructions the instructions are always in the context window. So imagine that every single time you interact this whole thing is in The chat window here right the files they are uploaded into the context window on demand directly or indirectly while the instructions are always there it's always as if you started chat with the instructions and that's important distinction right but now if I have the brand revamp project as I said it's for like an ongoing one bigger project I'm working on I could start uploading files you

know like my current product line uh spreadsheet you know I can upload my Current brand guidelines you know the colors the font the logos I can upload all of these things I can upload my sales data so he knows you know which products are performing better than others so I'm uploading the context giving as much as possible for this particular project and then I can start working that I will say hey um hey I need help updating my target customer avatar And now if I ask a very simple question in a structure of a prompt

that I wouldn't usually recommend because it's just a simple, you know, straightforward what we said here uh in our previous example having just a straight question, we're probably not going to get a good answer. But this time we have the master prompt, we have the system prompt. So all of the requirements for the prompt engineering I described before you know the the role the task the context the constraints the master prompt system prompt all of the Context is already there. So now I can finally interact with it in an very simple and straightforward way. You

know I need updating my target custom avatar and he reads the documents and as you can see it's very on point about the things that I included in my system prompt. Right? So whether I'm creating new assets, asking about specific products, asking about, you know, my previous brand voice and ask him how I should change it, you know, to appeal to More premium um uh customers, for example, right? So now it's getting very specific and the level of answers I get in this project versus me just asking straightforward the same question. Let's just see if

I turn on, you know, just goody auto and just say like, hey, I need to update my target customer avatar. So, I'm just going to get something like not very relevant to me, right? So, that is the power of actually building the Foundation for your AI workspace and that's that's where we kind of unlock the higher value from and leverage from from the AI. Now we are not done because now we talked about the projects but we have also access to custom GPTs and these are almost as your freelancers right there are there to do

specific tasks you know I say these are for doing and repeating so once you have a specific task in your business specific workflow you need to do again And again and it doesn't matter as much it's not evolving yeah you just need to do the thing so I think referring to this as freelancers is the most powerful. Um, we can actually create as many custom GPDs as we want for specific areas of the business. So here in project, we talk about brand revamp project. But for example, for custom GPT, we can have a listing strategist

AI, right? So let me show you how this works. So if I go back into my prompt Generator and I will copy this. Hey, I'm creating a custom GBD called listing strategies. Its role is to help me write, optimize and analyze product listings including keyword research, split test ideas. Help me write the system prompt and so on so on. Very simple request. So we will paste it here. And in the meantime as uh CH GPD is working on this we can go into this section into GPTs we will go to explore And then go ahead

into my GPTs. As you can see I already have my scaleboard chief operating officer but we will go ahead and create a new GPT. I recommend uh to go to the configure section. You know you can use a nice logo or avatar. You can name your GPD. So that would be listing strategist AI here. They could be a short description. Um optimizing listing performance. And now we have a few things here. Um instructions. We already know that this Is where the system prompt goes. We have conversation starters. That's essentially a button that appears at the

beginning uh of the interaction. So we can kind of pre-program it in a way that uh we don't just have to write something. We can have these buttons. So we can have create listing. There could be analyze listing performance, right? So when someone starts a conversation, they can click on these buttons immediately. We have knowledge. This is where uh it's similar with the project, right? Where we upload the project files. These are where we upload files that the custom CHGPT can get access to again on demand. It's not automatically uploaded to the context window GPT

can access them if we request them directly or indirectly. Right? So if I would have again my customer avatar here if he needs to create a listing he would refer to the customer avatar. But if he's analyzing listing performance Looking at data in a spreadsheet well he doesn't have to do that. He doesn't need access to the customer avatar. So he saves him some some time. So we have here the instructions and here as we can see uh for listing strategies here's your system prompt for the listing strategies custom GPT designed to fit seamlessly with

your master prompt context and ensure consistency. So as you can see he already in my prompt the builder he has access to my master Prompt. So he's generating this uh suddenly we have the role you are listing strategist MKL's expert partner for Amazon listing optimization. uh then objective maximize organic visibility, click-through rates and conversion rates through data back and keyword integration, conversion focus copy and so on. Then we have the context, the workflow, the output style, the tone and additional capabilities. So as you can see again, I would do a Little bit of refining before

I would upload it, but this is actually pretty good already. So I will copy it and go ahead into my instructions. And now every time I interact with this, he has access to um to this prompt. So he's operating again as we he would prompt him like this every single time. And I would also upload the files here, my master prompt. So he has access to it. I can potentially upload it here, right? That that depends what type of AI We are creating. If he really needs access all the time to all of the information

in the master prompt. So in some cases, I would recommend it. So he would have system prompt and the master prompt would go above that right very simple um it's in the markdown formatting right so it's beautiful everything is structured so I can open it and see the full prompt and uh now when we go here uh the as I said the knowledge files they can be uh uploaded Here uh let me let me jump in here so as we uploaded the master prompt we can also select the recommended model. I would recommend JG GPD5

thinking. I would also recommend code interpreter and data analysis that it allows us to analyze large data sets like your spreadsheets and Amazon report. And now there is one very beautiful function which is a little bit advanced but the beauty of the custom GPTs is that we can specify and basically program actions. As you can see actions here. It's different from the buttons here that are the conversation starters. I can go create new action. And as you can see this this gets a little bit intimidating. And we are not going to cover this in the

video because I'm going to create another video about this. But what this allows us essentially to do is specify give it some capabilities. And in this capabilities we can give it actions that Are essentially API calls. And long story short, what you need to know about API calls is that it allows you to interact with other tools and they have some nice examples. So if we go uh for example this uh this pet store what they have here we can give our AI some capabilities that are the API calls and one of them is get

inventory right and if we would test this it's essentially an API call to another software that returns data into our chat GPT Right so if I do this it's a little bit advanced but if you would do this for your business you can essentially get access to other tools and have API call let's say your inventory manager to send an API call to get an inventory report directly. So you don't need to go download the report, upload the file, and as you can see, it can basically get access to it through an API call. But

as I said, this is a little bit advanced. You don't need to do it now. But keep in Mind that this capability is here that we can specify as many actions as we want. Right? Very, very powerful. Um, I want to stress out that this may seem a little bit tedious to set up the structure. That's why I only recommend doing it for the large projects in your business, right? If you are like, hey, I'm coming up with a new title for my product, you know, or whatever my brand name, you don't need to create

all of this structure, right? This really makes Sense for areas for recurring high impact workflows and large projects like brand revamp, right? where you are going through multiple multiple steps on an ongoing basis or you have very well- definfined workflows which again you should >> [snorts] >> uh in your business that you do on a recurring basis and create custom GPDS for that. Now we are going to give AI access to All of your tools your Gmail Google Drive notion asana even your seller central this transforms AI from a chatbot into an agent that can

actually get work done in your business. We'll cover MCPS, model context protocol, which is how you connect AI to any software. You'll learn how to set up official connectors in cloud and chat GPT, and how to build and host your own custom MCP servers for tools that don't have official integrations. By the end Of this section, your AI will be able to read emails, search documents, create tasks, and pull data from anywhere you need it. Let's dive in. I would like to clarify the terminology we are going to use and the theory behind this. Don't

worry, I'm not going to spend too much time on this, but it's very important that we are on the same page and you understand these concepts. So, let's spend two or three minutes on this. First of all, you've probably used um KG GPT or cloud, you know, you your large language model. But if you are like me, the first time you used it, you probably had this thought like what if this could actually know what's going on in my business? What if this could actually go and do tasks for me, not just give me text

or images? And what if it could be also proactive to do things without me even asking for it? So this is where AI agents come in. It is essentially a large language model like your claude or Chat GPT which has the power to take action. It has the right tools and the right data to make decisions and it's also guided by a clear job description. But when we talk about giving tools to our AI so it can become an AI agent. Let me explain it like this. Right? Whenever we use tools such as your seller

central or your asana you know your task management whatever that is we are using a graphic user interface to interact with the code that powers the app right We don't want to write code you know to create a new task in your task management. You simply click a button. That's what we do. That's what we like to do. But when it comes to your large language model like Ced GPT or CL they can do it but they don't really like it right they are not good at it and it's quite clunky for them to interact

with a tool using the graphic user interface that was made for humans of course for this purpose there are APIs you know Application programming interface and this is nothing new this was created so softwares two different softwares can essentially communicate with each other using the code right so it's one program using the tools and data of another so we can give our large language model like chat GPT the APIs to call the tools let's say in our sana it can create tasks it can create comments and that's how we give the tools to our AI

but the issue is that every single software Whether it's notion seller central asana toggle whatever you slack all of them have their own API documentation where every single thing you want to do to create a task leave a comment requires a different API call which is a code and of course the large language model GPT doesn't know these we need to program it so it can do all of these things which is very very tedious it is possible and it works but it's very tedious to do this for multiple softwares for multiple Different tools right

and that is Why MCP is the future? That is the model context protocol. It is a standardized way to give the tools to our large language models because programming the APIs for every single tool that just takes too much time and to be honest I don't really understand how to do this for every single thing in my uh in my chat GPT or cloud. Right? So for this purpose an MCP was created which is essentially a server where your large Language model is connected to this MCP server and he can just talk to it and

explain what he needs while the server is taking care of the API calls essentially executing the tasks in that particular tool. So the lash language model can talk to the MCP server of let's say Asana and he can tell it that he wants to create a task or leave a comment and the MCP server handles the API call which of course is the code and we can then see that in our Osana he Performed an action. So the way we can think about this is that our uh MCP the MCP servers are essentially the user

interface for our large language models. Right? So this is the key understanding here why the MCP servers are so powerful because we can easily give a big toolkit for each of our softwares we use to our large language model and he can then do the things we do in that particular software and if we take a look at uh let's say Claude right they call these Connectors but these are official MCP servers we can connect our large language model to right we are starting with the with the basics, right? So, it's possible that you've already

used this, but it's important to understand all of this because we will build on this knowledge. So, bear with me when we take a look at let's say one of the connectors which is notion right here. You can see all of these tools and each of these tools is essentially the API Call. But we don't need to care about the code. We essentially just press one button in cloth to connect into our notion and the cloth then has the ability to do all of these things in our notion to search, create pages, update pages, move

pages, everything you can do basically uh in your notion and this applies to many many other softwares right so if you go to your connectors and don't worry we will set this everything up Live in this video uh many many connectors for different tools, right, for different softwares. You have Asana, Canva, you have just so many of these officially in cloud. Of course, I'm talking about cloud, but this is exactly possible to do in CHGPT. It's they don't have that many official ones yet, but there are new popping out all the time. And as I

teach you in this video, you will be able to do everything in Ced GPT or cloth, whatever you prefer. In Ced GPT, you would go into your apps and connectors, activate the developer mode, and then you are able to connect all of this as well. And we will now explain how all of this works. So now we will go to specific scenario where we will teach our AI our cloud to interact with many different apps in order to achieve a specific objective. So in this scenario I will guide you through this. Let's say you need

to prepare for a meeting with the graphic designer. Right? You have a Product launch uh of a new product. You have a meeting with your designer in 1 hour. you forgot to prepare, you know, and you need to brief him. All of the information is scattered. You know, something is in Gmail, something on Google Drive, there are some notes in Notion, and you simply don't have the time to do that manually. And why would you, right, when you have AI? I want to stress out that all of this I'm about to show you is done

live, right? So, if Something happens, my my PC explodes, right? That's that's what can happen here. But uh we will do this and if something goes wrong, you will actually see how relatively easy it is to fix everything. So what we want to do is to first of all set up these MCPS to connectors, right? So what I'm going to do is I'm going to open my cloud here. So let me just put this on the side. And here I logged into my cloth. For these connectors, you are going to need a Premium account. So

you can you can set it up. Then we will go here into uh the settings, right? There's my system prompt and everything. And I will go simply into connectors. And as you can see, we are now here. Google Drive, Gmail, Asana, notion. If you cannot find it, you will find here uh a list of all of the official ones, connectors here that are very very easy to set up. And now I'm going to connect to all of these Gmail, Google Drive, notion, Asana. What This means is that we are giving Claude access to these tools.

So we can then do everything uh I described to prepare for the meeting and to access all of this. So let me now set this up. And as you can see, connecting the MCP to these official ones is very easy. I'm just going to log into my Google just share the permissions uh and we'll continue. And now we will wait for it to load. Now it's connected, right? As easy as a few Clicks. Now I'm going to continue do Gmail uh notion and asana. So let's do it now. So now we are ready. We connected

our Google drive, Gmail, Asana, notion. It is literally just a few clicks to do this. That's why we are starting here because it's the easiest way to set up these official MCPS. And now we are ready to actually prepare for the meeting. Right? This really this example this scenario the purpose is to show you how it can interact with many Different tools and basically pull everything you need. So I will hop in here and start a new chat with cloud. And if you click here you will then see that it has access to these tools

right web search of course that's basic but then we also have the Google drive gmail we can potentially connect our Google calendar right asa and notion. And when you click here you can see all of the things that it can do within that particular tool. So let's begin. First Of all, we know that somewhere in our email, there are emails from our supplier about the product specifications, right? We want to launch this product. Uh I don't remember the size of the box. I I don't remember anything to be honest. So I'm just going to ask

Claude to fetch this for me. All right. So we asked for the first step, which was extracting product specifications from email. So as you can see, I just uh wrote a simple prompt. Ideally, I would spend a little bit more time trying to be more specific, but it still works in this example. You know, we're launching new product. Martini glass is set to four, getting ready for meeting with my graphic designer. You will be getting information. Uh, fetch the details, blah, blah, blah. First of all, go to my Gmail and check for all emails from

our suppliers regarding martini glasses. There should be our product specs somewhere. As you can see, He searches, you know, he found two emails uh packaging, final product specs for Q4 launch. So, he found it. I've grabbed the product specification and packaging details from your supplier emails. Beautiful. So now we can go and our next step is to find competitor analysis data on our Google Drive. Right? We we gathered reviews from our competitors and now we want the insights that will be important when creating our listings. Again, I don't know where that File is. I don't

have the insights. So let me just ask and see what Claude will find. Now we of course cut in this video so you don't have to wait for the AI to perform the steps because it usually takes around minute or two to do this but the next step to find the competitor analysis on our Google drive you know now go to my Google drive and check the product development folder that's what I remembered that's it's there and once you find it analyze the product reviews From the document uh the insights will help us position our

product better targeting their weak points uh so of course you can see how what he actually does right he's searching he found the product development folder He finds the file, you know, he fetches the file and then he already analyzed the reviews in the file to find the weak points. Right? So now we have it, he has the information. So the third step is actually we had a call in this example With um with our brand manager talking about the strategy for this product that will also be relevant. And now I want to ask Claude

to go to notion where I know that there is a transcript of the call and ask him to just summarize the key points from the strategy that might be relevant for the graphic designer. So I'm going to send the prompt and see what happens. All right. So now we ask him to go to notion find meetings. We had a meeting with our brand manager About the strategy find the document read it and remember relevant information for the graphic designer. So as you can see he connects to notion finds the right database finds the right document

reads it remembers the detail and so you can see that I'm not making stuff up. Uh there is an email you know there are multiple emails about the uh the product specifications. There is a document on the Google drive and we have also in our notion database there is a Transcript right so he actually reads the transcript as you can see it's just a transcript he reads it and identifies the key information from uh from that so perfect now we kind of want to gather everything into a brief he want we want him to create

you know just some PDF we can then show to the graphic designer some one pager so I would ask for him to create it so we will use it during the call with our graphic designer as you can See, we ask him to generate a simple Onepage PDF where we summarize the product specs and essential information for the graphic designer. And this is what why I love love cloth because he's able to create these files. You know, it's a PDF I can download and continue sharing uh right here. So, let's take a look at it.

Uh well, it's not one page, but still pretty good, right? We we have the information, the product specs, everything that's important for the graphic designer. the brand direction That came from the meeting. Uh the target customers, we talked about competitor weak points to address that came from the reviews and design deliverables needed. And this is beautiful because he already kind of knows what the let me if I can zoom in. Um just read it to you. Uh the design deliverables, he understands Amazon listings without me explaining like we will talk about systems prompts and everything

but he knows we need main Image, lifestyle images, infographic A+. He specifies the deliverables without me explaining this and it's perfect. He understands Amazon uh better than than most sellers, I guess. Okay, so now we have everything we need. We just need to put it together into an agenda in notion, right? Because we want to get ready for the call and we want to put it into into notion. So now I will send him the prompt to put everything together and help me prepare for the meeting. And Lastly, we ask him that we want to

create the page in notion so we can prepare for the call as I mentioned. So as you can see, he went, you know, finds the right database. Uh he actually even looks up what other uh pages in the database look like. So he can, you know, replicate the formatting and then he creates the page in uh in notion. And if I go into the database right here, you know, uh he creates it. He also, you know, meeting purpose, brief Alex, Graphic designer. Uh then we have the key documents. He links everything the PDF you know

the competitor analysis file so we can take a look at it uh and the agenda you know and of course it's very detailed as I mentioned he understands all of the context we we shared but he also understands pretty well without any other prompting like what Amazon requires you know maybe there will be some small discrepancies but at the end of the day it is a huge Huge timesaver you know and here's key messages to visualize you know rainforest stem prevents breakage now He's referring back to the negative reviews from our competitors. And he also

creates like a timeline uh that we might agree on, you know, questions for Alex and success criteria, you know. Uh all of this is here. And now I will show you what that would look like. Let's say the meeting just finished. And we actually want to um Have a transcript here. We could have another setup to actually track uh the meetings. So we would have the data right here. You know, I just created uh a transcript what that could look like a meeting with the graphic designer. So I will copy it here. Just put it

here as a page. Of course, I wouldn't do it manually, but that's graphic designer meeting notes or transcript script. Um so of course this would be Recorded on the call, but of course I'm just demonstrating this so we don't have a real transcript. But what we can then do if the transcript is here I can actually go let's say after the call is finished and I want him to go to Asana and create a task. So I would simply go you know let's say I just start a new chat and I will send him requests

to find the meeting uh everything that was discussed and kind of push it into Asana as a task. So that would be the last Tool in this demonstration. So now as you can see we ask him to essentially do two things. go to my notion to my meeting database, find a transcript of a call with my graphic designer Alex, read the transcript, clarify all the steps and then push the task into Asana. So he does that, right? He goes to notion, reads the transcript, identifies the key steps and then he pushes that into Asana. He

creates the task and as you can see sometimes he makes some mistakes But he corrects himself. You don't have to tell him anything and it this took like two three minutes, right? So you can just basically do this almost instantly uh even multiple steps within one prompt. And if I then go into Osana, you know, I can see that uh the task is here with the details. I could then ask him, you know, to add the subtasks uh so we can track them accordingly with the deadlines. But of course here you get the point. So

now we covered the first Category of the MCP servers. these official MCP connectors that are built-in integrations available directly in your cloud. You just find it in the list and you click a button. That's extremely easy. And of course, as we go, there will be more and more of these. But you might find that sometimes there are not these connectors for things that you would actually like to use. Maybe there's a tool or software you really want to use, but somehow it's not in the Official connectors. So let me show you the next category. Official

provider MCP servers. Those are official servers made by the tool providers that have a link. They are online but they are simply not in cloud. They are not in the connector list and you need to add them manually. Which brings me to the second scenario. Let's say we want to do a competitor review analysis. We want to scrape reviews directly from our cloth. So we will need some website scraper to do That and we will use appy. And basically what we are trying to do is to scrape the reviews from cloth let him analyze that

and then generate a nice report we can then share with our team and u you know discuss further. So I chose Appify and as I said here if we go to the connectors and I would browse connectors there is no such thing as aify in these web connectors which means that I have to do that uh manually right so EPify is one of the website scrapers so if me let Me just open this full screen and we will go to aify the first thing you need to do is to actually set up your account right

so aifi's website scraper. You can scrape many things. Tik Tok, Google Maps, Instagram, but also Amazon, right? You can scrape Amazon reviews. And so you will log in into your um into your account. Again, very simple login where we uh where we choose our account and what we can do actually they provide for free quite a lot of things we can do, You know, even on the free tier. So that will be more than sufficient. and then set up our appy MCP server right so we will go to MCP ampify and as I said this

is an official MCP server for this particular tool right and I can find it here this is a setup again now I understand it it might get a little bit intimidating but don't worry I will explain everything in detail so what we are going to do is that we will add these actors which essentially on epify Those are uh The way is how we scrape specific things. Like we want to scrape Instagram. I would add Instagram. But I want to do Amazon review scraping. Um, and I believe I liked this one. So I will save

it. Right now I have uh everything I need here. The the scraper. I will scroll down and they generate a link. That's the most important thing. Right. That is the link to the MCP server. As you can see that is the endpoint to the integrations. So again, It's not intimidating, right? If we just copy this address, then I will hop back into uh into my cloth. I will go add custom connector. I will go epify MCP. I will add the URL which already includes that particular actor which is basically the way we scrape you know

Amazon. So we don't want anything else right now. So we will add this. And now I have it here. I need to click connect which is basically the same thing we did Before with notion and so on. We just had to add the extra step of adding that URL. Right? So I will choose me allow access and now it's active. So now we are good to go right as you can see it was quite simple. It didn't hurt and we are good to go. We connected to clot took 30 seconds. Right now let's say this

in this scenario we want to scrape reviews. In this case I would probably create a custom project with a system prompt that To say you know you are an Amazon review analyzer. Your job is to do x y and z. So you don't always have to refer back you know with every single prompt. It's basically the same thing as in chat GPT. You can create these uh these projects right? So you create a new project. But for this example I'm not going to do that right. I'm going to simply take the ASIN and ask uh

if he can scrape it using the EPify MCP and let's see what we will get. Okay, so I asked him to use The EPI MCP to scrape Amazon product reviews on the Amazon US. As I said, I would use probably a bit more comprehensive uh system prompt, but this still works. You know, I gave him the as scrape and analyze all of the insights from reviews. I actually forgot to specify to use the specific actor I chose because I like that one the most. So I just then refreshed it and asked him the same thing.

And as you can see, he used Empify, you know, went to Amazon, got the results, uh, martini glasses, 340 reviews, five stars, what works, the problem, uh, quick takeaway. All right. So, as you can see, he gathered the key insights uh and he has the information, he has the the reviews and here I also uh said that we are going to generate a report, a nice dashboard, right? So, let's see what we can do with Claude. So, as we asked uh Claude to use the data he got from Appify, right? All of the reviews to

Create a nice interactive dashboard. So, as you can see how he thinks about this, right? He goes back to Appify to download the full report and then he starts working on a nice uh dashboard. So let me actually put this full screen because I really want you to see this. When you go view the dashboard, we have the data we just gathered in a nice interactive dashboard. Again, if we had the system prompt, we can like um unify this. So every single time it's the Same, right? So we have top positive main concerns. You can

see the uh the split of the reviews um reviews over time. Uh then we see you know customer sentiment by aspect, the quality, appearance, size, sturdiness, thickness and weight, you know, especially the thickness, right? We talked about this review explorer. It also gave us uh and it actually allows us to to filter out the reviews. Uh he he downloaded as well, right? And it's an interactive Dashboard. There are some key takeaways and so on. So we can definitely then play with this a bit. But now you can see the power of what you can actually

do everything from cloth. We gave him the ability to scrape the reviews, create a dashboard, everything from one place. So epify was an example of an official provider MCP server that's not in the connectors. Now we will talk about another one which is an MCP gateway service. This is essentially a Third-party platform that gives you access to some tools that they might not even have an official MCP server, right? For example, some of the smaller tools like let's say Amazon seller central, they do not have an official MCP yet. I assume there will be some

in the future. But what we can do is we can actually use something like an MCP gateway service for example Pyream which I will show you right now to get access to many different MCPS essentially through one Thing right. So if you go to something like um to something like pipedream Zapier does the same thing but uh piped is good. We will go to product MCP. You will set up your account. You know, you will um you will log in. So, I will go continue with Google. We will log in. Again, you can use this

for free. And now we have the MCP pipe dream. We see all of the MCPS that are there. And as you can see, there are a lot of tools compared to the connectors we had nine 2,900 to be to be specific compared to like 50 we currently have in cloud. Right? So these are some of them might be official but uh actually here sorry actually here it's uh these are developed by pipedream right those are the custom ones they developed for some of these tools so let's say the one for notion will might be different

from the official one right uh but what we can do then if we want to connect let's say slack so you search for slack um you Click on it and you will see the available tools what this particular MCP remember the tools what it can do and as you can see this one can do a lot of things. It can do pretty much everything we need. Uh and if you would search for you know Slack MCP you might find that some of these actually with Slack I do not know right now but some of these

simply do not have an official one. So what you can then do is that as we use the Appify uh URL you first go to Pipeream you connect the accounts you want. So if I would go to you know connect Slack, I just continue uh connect it whether it's you know authorization of the account or the API key it's very simple and then I would be using this one link to access multiple other ones right so this is the gateway you connect one you get access to many tools right so as they say here um

10,000 tools right 2,800 APIs 10,000 tools so as you can see then then you Have one link you can connect Google sheets Google can Shopify anything you can think of and of course even our favorite Amazon. So if we search for example for selling partner you will find uh that there might be some for seller central. Of course these ones as I said are not official so they might be a bit limited but eventually they get better and better and there are actually better ways to do this. uh I will explain in the third scenario

but As you can see there are some uh available tools you know get order details generate sales and inventory report. So if I would connect this with my account uh use this URL you know to add it to my clot as I did with uh with appy I would get access to these tools so I can just call my sales reports through clot and tell him you know put it in a nice dashboard compare it with previous month you know go to my Google drive and you start combining all of These things into something extremely

valuable right so this is piped uh gateway but as I said it's likely that more and more of these tool providers will develop their own official MCP servers. So it's the same like epifi or even the official cloth connectors. So you just go click click click and you are good to go. So now we have the gateway out of the way and we are getting into the third scenario and this one is the most advanced. I don't want you to get intimidated. Just follow the steps right. It's relatively easy if you follow this step by

step. Plus, if you have Claude by your side, he will explain everything you need uh to do that, right? So, if you can follow instructions, you will be able to set this up. We are getting into community MCP servers. Community MCP servers are servers that are not official, but people shared them on uh GitHub, where you can essentially use The code of the MCB server and host it on your own computer. It all sounds a bit maybe intimidating but don't worry I will explain everything in detail because in scenario three we want to take it

up a notch right what we want to do is to set up a proper automation because what we've done so far it's amazing but still it's way it's very reactive right we just need to send message to get something what if we want this to happen without our without our Prompt right without us on a regular basis we need to set up an automation and for this we can use N8N but If you are like me, building something like this is, you know, uh, quite scary. I don't know where to begin. Uh, and what if

I could use Claude to actually build this for me? What if I could give him access to N8N to do this for me? And of course, why I'm showing this, you can do that. So, we will do two things. We will demonstrate two things. Setting up a Self-hosted MCP server. In this example, it will be an 8N. And I will also show show you how the how Claude then can use the N8 and MCP server to generate an automation specifically an automation to monitor your competitor prices you know daily across multiple Amazon listings. I want

to build a custom automation that tracks prices that exchanges and alerts me when someone undercuts us. Right? That would be beautiful, very useful automation we can set up. Of course, you Can then create whatever you want, but this one would be, I believe, useful to most people. So, if we go to piped, let's say we would want to do it through piped, right? We said uh that we can use this gateway. So, why wouldn't we do this? Why even talk about self-hosted? N8N is right here, right? No. Uh if I take a look at available

tools, uh there are no available tools for this app yet because N8N is quite complicated and uh they are simply not On Pipeream. There is not a good server. the best server. If I actually search, there is an N8MCP which is very well rated and works very well. Um, it's not official. Someone, you know, programmed it himself and he's now sharing the code on GitHub. So, if I go here, he's sharing everything right here. He tells you how to set it up. And again, might be a bit intimidating to do this. you know, you don't

know where to begin, but if you follow the guide, it's Relatively simple, and I will show you in this video how to how to do all of this. And then your claude will unlock this amazing thing, work with N8N. But that's not just the sole purpose. You will unlock the ability to self-host MCP servers, which will then basically give you the ability to control whatever you want and not rely on official MCP servers that might not be as good or they might not be even available at all, right? And this is how you get ahead

of The curve because this is the stuff that no one is going to do right and we will then talk about you know building your own MCP server as well but in this case we are going to use something that someone else built. So for this what we need to do is we need to use cloth desktop. It doesn't work with the web app if you are self-hosting it. Right? So you will just go to cloudai/d download and you would download the desktop app. There is nothing special About it. Right? you just download Windows or

or Mac OS and you are good to go. The next [clears throat] thing is to set up your self-hosted N8 and MCP server. So, first of all, this guy who created this basically did all the heavy lifting for us. So, as you can see, there is a prerequisite to have NodeJS. So, you would click on the link here and what you're going to do, don't worry about it. Just download it, install it, right? So, if you're Windows or Mac, Right? So I already did this. You would download this file, right? And install this that will

allow you to then self-host the server, right? So as you can see, you just download it, install it. Nothing special about it. What then you'll do if you are on Windows to be honest, uh on Mac it will be slightly different, but I don't have Mac. So you will have to find this step somewhere else. But if we go to the command prompt uh and run it as an administrator on Your computer and you just run this code he has here. So I will click here to copy right and enter. And now my MCB server

for N8N is running. That's it. Now I have to connect it to cloud right so just uh I will probably write it in the description if you have Mac it will be slightly different but at the end of the day it's exactly the same thing right you install NodeJS You run this code it will download the the code that's on GitHub essentially and runs the server the MCP server with all of this code behind it so you can start using it now my server is running here you know server startup completed and now I can

go to my clot and I need to connect the clot and give him the ability to actually uh get access to um to this MCP server and as you can see it's just this configuration here again it's a code I don't know what it means But I don't care because it's simple to put it in there right what I will need is two things my N8N API URL which is basically my N8N instance in order to do this you will need the premium N8N account but that's kind of uh uh understandable right so if I

go back here I will go to N8N you know this is my um right here this would be the instance just the URL basically you get and then the API key and I will also show you how to get the API key from N8N So if I go back here and go ahead and turn on cloud so now in I'm in the desktop app right it's important it's exactly the name just gives us the ability to go to settings and then go to the developer which we do cannot do right as you can see local

MCP servers add and manage MCP servers that you are working on. So I will go edit config again. Very simple. Just you know slow down the video, watch it a few Times if you need to. You go edit config. Now it finds the cloth desktop config JSON. Right click and you edit it in your notepad. So you can edit the code. And as you can see currently it's empty. There's nothing because I don't have any local MCP servers. But the smart guy here, he tells me to use this full configuration. So I'll go copy back

to my notepad and I will copy all of this there. Right? [snorts] Simple. Now we go to API URL. As I mentioned, we need where is it? Here we need the N8N instance. So that would be my website link. So I will just replace it. Be careful not to mess this up. So I will give the the instance here. And now we need the API key which is secret. You shouldn't show it anywhere. So we will probably blur it here. But if you go to your N8N your account, you know this is your dashboard. You

go to settings N8N API Create new API key and expiration. I don't want this to expire and I will say cloud on cloud desktop 3. And now we will save it. And my API key will appear. We will blur it. So here I will just click to copy. Now I have the API key. I will go here, highlight it, and copy the API key there. All right, make sure that's the whole code. Save this file. All right, I'm basically done with this. Um, then I go back to cloth and I need To turn it off,

right? So wish me luck that everything works as intended. Now I turned it off and I need to turn it on again. And now when I go to uh settings and I go back to the developer, you can see N8 and MCP running. We are connected to our MCP server that is selfhosted running on my own computer. So just to check you know I can then go ahead and say use the um N8 MCB server to do a health check just to see that the connection is running well and we Are good to go. The

it's healthy it's connected everything works as intended. So now uh let me switch back to my notes here. We connected the the MCP server and some of these MCP servers that are a little bit more complex like N8N the pro the developer who built this he also provides a system prompt right we talked about this creating a project with a system prompt uh to make it even more powerful so the AI knows how to interact With the MCP server because it's so complex there are so many tools so you can then copy this system prompt

you know cloud project setup I copy this system prompt and I would go to projects and I already did this and I then workflow generator and you go into instructions and you copy this. So he knows how to act. He knows how to uh do all of these things, how to interact with the MCP server. And I already did this, right? That that makes it that Much more powerful. Very very simple to copy it there. And now if I go here, basically the next step would be to explain cla what kind of workflow we want.

And here you want to provide a lot of detail. you want to be, you know, as descriptive as possible so it can be uh accurate. But here I already did this because it actually takes a little bit of time so I can just show you the final result. So what I did as we talked about this that we want a competitor price Tracking. So I used this project the N8N workflow generator and wrote this very simply as I said you should probably include more detail so it's more accurate but I just said help me create

a simple N8N automation for competitor price tracking on Amazon. It runs daily, uses Appify to scrape us Amazon for our top five competitors. If it tracks their prices day by day in the Google spreadsheet, if there's a price change, we will get notified on Slack. So, it Goes and does all of these things. It takes a little bit of time. And by the way, highly recommended to use the other model opus for this, the smarter one uh for complex challenges because it is a lot of work for Claude and it will eat your tokens for

the day. But it is definitely worth it because it saves so much time. Um that he creates this, right? He created the workflow. I will show it to you. And he also because the system is super smart, he also prepares A setup guide. It literally tells us what we need to do to finalize the workflow because we need to connect the accounts, you know, and it's still quite intimidating, but he creates this beautiful stepbystep document that tells us how to finalize the workflow to make it, you know, to deploy it into our business, you know,

required accounts, Google sheet setup. He even tells you what the Google sheet should look like. Just copy paste price history, you know, He tells you epify store, you know, search for Amazon product scraper, try for free. He tells you everything in here. It makes it so much easier. And so I have these instructions. And if I go back into my N8N and go into my workflows, this is the Amazon competitor price tracker he created. So now I have the full workflow. And as you can see, it's not finalized because he tells me here uh how

to finalize it, you know, how to uh how to set everything up. But As you can see, it's not only the notes. He even formatted, you know, what the message on Slack will look like. And when I'm looking at this, I wouldn't be able to to write this. Like I I don't even know what's what's going on here. But if I follow the workflow, uh this would work, right? This would basically do the job. If you know, if there are some tweaks required, uh here he has, you know, the the correct API scraper already linked.

You know, Everything is here. You just need to fill in the asens. It also knows that with the epifi scraper you need to wait for the scraper to get the results. So it asks the scraper then it waits and then it goes back to look up the data. I don't know how it knows that but it knows and get competitor as so he says your Google sheet ID right and this really corresponds one to one with the document it created here. So as you can see it even makes it easier to actually Create your own

MCB server create your own automations. So if we just recap, we were able to generate a new workflow. We were able to set up our self-hosted MCP. We understand MCPS in general and it unlocks so much with your cloud or chat GPT that as I promised you, you are now ahead of 90% of people. I talked to dozens and dozens of sellers and I guarantee you no one really is is doing this. Right? I want to stress out that with the N810, it's not perfect. you Still need to learn, you know, to to utilize it

fully. Takes a lot of usage, as I mentioned, it eats up the tokens, but extremely powerful. And a little cherry on top, which is basically the most advanced thing uh here is that you can build your own MCP servers and host them online, right? So, you can create your own specifically for your business, connect all of the tools and essentially create your own MCP gateway with our MCP servers. And that's what we are doing Right now and experimenting with that because you can then pretty much create whatever you need and plug in your claw chat

GPT into it to do incredible things. Now we are going to install your playbooks directly into AI. Instead of explaining your process every single time, you'll teach AI your exact procedures once and it executes them perfectly every time. We'll use cloth skills to do this. Skills sets you turn Your SOPs into AI specialists. Whether that's a listing strategist that generates listings following your framework or a review analyzer that pulls insights exactly how you want them. And while this feature started with Claude, it's setting the standard for every AI platform. By the end of this section,

you'll know how to build your first skill from scratch and create AI specialists for your most important workflows. Let's begin. So why are we Even talking about procedures for AI? What is it and why should you care? If you've ever used a large language model before, like cloth or chat GPT, you know that it's not always easy to get consistent results. That sometimes it gets it right on the first try. It's perfect. And other times you need to go back and forth multiple times and it still doesn't get it quite right. And this becomes a

problem if you want to standardize quality in your business, Especially if you have a team and everyone is using AI differently. And these are actually the exact same problems we face with people if we don't have procedures. You know, they do things differently every single time. They need constant back and forth to explain what the task is about and they struggle to maintain quality and consistency. And for this reason, Entropic, the company behind CLA, introduced clot skills. They allow you To do the exact same thing as procedures. They let us apply a level of structure

and consistency to AI. So, it also allows us to then train the AI on our frameworks and on our workflows. So, we moved from kind of generalist AI assistant to an expert level AI specialist that can deliver these predictable results. So what we are essentially going to do is that we are going to use cloud skills to basically install our SOPs into our AI. So then The AI executes exactly as we designed it. And since this is designed by Entropic, we also know that Entropic was the first one who introduced MCPS. Those are the connectors,

right? Those are how you connect your tools like cloth or chat GPT to any other external tool and it became the standard. So I believe that skills will become a standard for process execution. So for now we only have it in cloth but sooner or later I believe that we will see that in any Other large language model. So before we dive into some explanations I actually want to show you how cloth skills work. So if we hop into our cloth and go into settings in order to activate this we will go to capabilities and

then we need to make sure that code and execution and file creation is activated that will enable us to use skills. Once we do that you can scroll down and find this section called skills. As you can see repeatable customizable instructions That claude can follow in any chat. There will be multiple skills by default and these are there by entropic right we have some brand guidelines canvas design mcp build builder theme factory these are more or less just to show you how these skill works but the most useful ones will be the skills that we

actually create right the skills that we create based on our SOPs so let's create the first one a simple one we will use uh a skill to essentially help us create SOPs For our business. Since we are already talking about SOPs, this will be a simple example. So let me show you. So now we ask Claude to use its meta skill which is a skill creator to create an SOP generator skill. Right? We are asking Claude that every time we upload a transcript of an instructional video to turn it into a standardized procedure. So now

we are creating the skill. As you can see, we will say that the procedure should include objective, Inputs, highle summary, and step-by-step breakdown to keep it short. And we also asking if there is anything else you need to know from me in order to create this skill. And now you can see that Claude is actually accessing the meta skill, which is the skill creator, which is a skill to create skills to build our SOP generator. And as you can see now, he's asking us a bunch of questions. what kind of uh format the transcript will

be, what kind of output we want, uh The tone, audience, step detail level and so on. So now let me answer these questions. So we simply answered all of the questions. As you can see now we are standardizing this particular procedure right this workflow is there to help us create SOPs for our business. Uh as I said this is a very simple example but since we are answering these and specifying these we will make sure that every single time we create a procedure for our business it will be following These particular steps and it will

be in the same format every single time. So as you can see Claude used the skill to generate this file right we specified what our SOPs based on the transcript should look like and it will have objective inputs highle summary and now we have this file which we will now download and the last step we have to do is to go back into our capabilities here and simply upload this file into our cloth. So now I simply added the skill And as you can see here we have our SOP generator that converts instructional video transcript

into standardized operating procedure. So let's try it. Let's try it in the chat and we will say you know hey Claude I just added the SOP generator skill. Here is a video transcript and here actually I have a simple transcript of a video I recorded before. So I will copy the transcript from uh Loom and copy paste it. Right. There's no other explanation needed. Now, every single time I want to create an SOP, I simply ask Claude to use the SOP generator scale. So, let's see. So, now as you can see, Claude used the transcript

to create an SOP. So, it includes everything we ask for the purpose, when to use this SOP and so on. But, you know what? I realized that having the SOP here in chat is not great. I want to save it in our database in notion, you know, in our playbook library and I don't want to copy paste It. So let me go here and actually update the skill right in this chat where we ask Claude to create the skill uh we will ask him to add an extra step to take this SOP and save it

directly into our notion using the notion MCP. So let's see. So we simply ask Claude that we would like to add one more step to the SOP generator skill. The last step should be to save that SOP into our notion in the playbook library database. And as you can see, he simply created a New file. So once again, we have to download this skill and then include it and upload it back into cloth. We simply uploaded the new file. So now our SOP generator has this extra step to save the SOP into our database. So

we will try it in chat again. Copy paste the transcript and see if all of this works as intended. So the only thing we had to do is to tell Claude to use the new SOP generator skill and copy paste the video transcript and he will follow the steps Step by step every single time. So first he reads this the SOP generator skill what it's about. Then he finds the notion database and then he creates the SOP based on our previous instructions. And as you can see he automatically saved it into our playbook library you

know troubleshoot inventory forecast accuracy. So we have the objective we have the inputs we have the highle summary and we have the stepbystep breakdown. And the beauty of this is That every single time we now create a new SOP, it will have the same structure and it will go directly into my notion. So now you can see the power, right? We essentially created again a meta skill for this for creating procedures because every business should have procedures with unified structure. And as you can see, I would probably spend a little bit more time to update

the skill to make sure that it look exactly as I wanted to wanted to. So then in the future, I can Always create it the exact same way, right? And no description needed. Every single time the output will be the same. So now that you see how you can create cloth skills, let's talk about the types of cloud skills. Because the first type are the simple skills, right? They are kind of guiding the thinking and the structure for the AI. And this essentially is almost like a modular prompt, right? It is mostly textbased defining the

tone, format or just Specifying the logical steps. But where it becomes powerful is when we include analytical skills because we will be essentially guiding Claude to analyze, calculate and visualize data every single time the same way because this is important. If you have your reports in business, you essentially want to analyze the reports every single time the same way, right? Following some steps and following some operations and calculations. So these analytical skills Will trigger the cloth's built-in Python environment. This is essentially code that allows him to interact with large data sets and help you run

an analysis. And we will take a closer look at all of this. And lastly, we have the connected skills where we essentially one of the steps in our skill is to interact with an external tool which is what we did right now, right? We taught Claude to use our notion, right? So through the connectors, I actually have another Video about this through the connectors. We connected our notion. So now he has access to it and we didn't have to specify it again, right? we just told him in the scale and now he knows every single

time he creates a procedure he needs to save it right here. Uh so those would be the connected skills and of course you can combine these right because your analytical skill can also be a connected skill interacting with other tools. So maybe you ask him to First get the report from your Google drive then analyze it and save the results in notion. Right? That is the beauty of it. We can predefine the skills in any way we want. And as I said, if I was doing this for real and not just for a live demonstration,

I would spend a little bit more time. You know, that's the idea. We spent one hour, two hours building something that will save us the time later on. Now, in terms of the skill sources, you've seen The official skills, right? Those were the brand guidelines by Entropic. And these are kind of just to show you what's possible. The powerful ones are the custom ones. You know, those are the personalized SOP style workflows, but there are also partner skills. Let me show you. If I open this website by Entropic on Google on GitHub and scroll down,

there will be partner skills. And since this is very new, right, launched like two weeks ago, you can see that There's so far only one and that's notion. And if we open this, we actually get a database of skills. You can download skills someone else created and you can download them and upload upload into your cloud to get these capabilities into your cloud. So as you can see notion, the team behind notion created these skills for meeting intelligence, research and documentation, knowledge capture. All of these skills you can just download and Upload it into your

cloud very easily. And of course with that you also get the community skills which are the custom skills created by other people. And as you can see already there are marketplaces for this uh and as I said this is this is fairly new. So now we have a even marketplace where you can browse these you know agents and automation and you can see what other people created and if you like it you know there is of course some security Make sure that you are actually downloading downloading it from a reliable re from a reliable source

but if you download it you can upload it into uh your cloud and start using it. So you don't have to just create it yourself and I believe that in in the future people will create powerful skills you know for e-commerce and Amazon that you will be able to upload you know to analyze specific type of reports and perform specific actions. So You are essentially turning your cloth into the specialized agent. But now we are talking a lot about prompting and skills. So let me explain the structure behind this because we all know that your

AI is only as good as your prompt. In other words, garbage in, garbage out. And in order to really use all of this properly and really have the real leverage here, we need to understand this. First of all, the components, roles, and purpose, right? You've heard About master prompt, you heard about system prompts, MCPs, skills. So let's just break it all down and see how all of this can work together. So of course the first component is your large language model, your cloth, your chedg that would be the brain that makes the decisions, reasoning and

all the interpretation, right? I personally prefer cloth right now but it can be chpt. Then we have a master prompt. This is your identity, the user's identity, Right? Defining who you are, your perspectives and goals and your communication style. So for example, if I go into my cloud into settings, this would be my master prompt, right? Explaining who I am, what are my goals and so on. We need to feed the context into the AI to really use this as much as possible. Then we have the project. That would be your system prompt. That is

where we are defining the role of the AI, right? It defines who the AI is, the Function, responsibilities and boundaries of the agent. So if for example I go into my projects here I will have my data analyst and as you can see the instructions here are your specialized data analyst for Amazon and e-commerce businesses your expertise includes PPC campaigns and so on. So here I'm describing who the AI is right that is the role the system prompt and then we have the skills right those are the playbooks the as I said modular Prompts those

define how task should be done the SOP logic best practices your frameworks and your standards and lastly we have MCPs the model context protocols and this essentially allows us to connect our AI to other tools because we don't just want to operate within the limits of our chat we want to interact with the world ideally through the skills. So we can connect our Google sheets, we can even connect you know uh Amazon potentially notion as sana Everything you need can be connected through MCPS and now if you connect all of this you will really unlock

superpowers and 90% of people really don't use this properly. So let's take a look how to actually implement this the right way into your business. So if we take a look here at the real world application because here it really becomes powerful. Now imagine every single team member on your team is having their own AI co-pilot like this. And I'm going to show you some specific examples and the AI co-pilot is trained on the exact SOPs and workflows. They have access to all of the tools, right? And most importantly, they have all of the relevant

contexts. They know the AI knows who it's working with. the AI knows what it should do because it has its own system prompt and of course it has access to the skills. So it's only waiting to be prompted uh by the user to do the right job. So in this example, Let's say we have a brand manager, right? Most of us have a brand manager in our business. So the brand manager, the master prompt would define that I'm a brand manager for this brand. These are my goals. These are my responsibilities. You know, this is

the company I work for. These are my KPIs, you know, and these are my preferences. Talk to me like this, you know, I prefer robotic, straight to the point, whatever. And also the KPIs, what Matters to me, how I'm rewarded in my job, right? What I'm what are the numbers in the business I'm responsible for? That would be my master prompt. Then I would create projects, let's say like a listing strategist that I'm going to show you today, right? The listing strategist, uh, let me show you here my listing strategist. Uh, it's a system prompt.

there are listing strategies for Amazon sellers. Your goal is to maximize listing performance through datadriven Optimization and then we have a bunch of uh other instructions on how to behave you know so now the AI knows exactly what its role is in a relationship to me to the brand manager. It will help me create optimize and analyze Amazon listings you know this acts as the system prompt. In most uh large language models they call it a project but uh essentially it is a system prompt instructing the AI how to behave. And lastly, we will have

the skills. Right? So in this example, I'm going to show you three skills. Listing creator, you know, generating keyword optimized copy using brand guidelines, review analyzer, which extracting patterns and insights from customer reviews. And also to show you the analytical capabilities, an SQR analyzer that evaluates performance data and identifies improvement areas. And also then the MCPS because we kind of want to connect, you know, to our database in notion. uh we will use EPI Which is a website scraper to get access to reviews uh to the content of the reviews and potentially Google drive you

know where we have our product briefs and listing archives. So let's take a look at it. Uh first of all when creating the skills that is of course the first step. I like to keep it you know under one roof in a skill create project you know dedicated workspace for creating and refining custom cloth skills because as I said sometimes you Go back to make adjustments to the skill as I did with the SOP generator right I asked hey it would be nice to actually save it in notion. So I like to keep them organized

like this in a project called skill creator. And of course he has a system prompt. You are a skill creation specialist. your primary job is to help me create uh high quality custom skills for Claude. Now, the first one will be Amazon listing generator skill creation. And I'm not going to do this Uh live because it's actually quite a lot of work to do this, but I will explain everything I did. So, in this example, uh we wanted to create a listing creator. So first of all I copy pasted a specific listing into claude to

help me create this skill right. So I said this skill will be generating a full Amazon listing title, bullet points and ideas for secondary images and A+ content. It will generate this based on a list of keywords that we gather our During our research phase. Right? So we are already prompting it what will be the input and what will be the output. The input will be a list of keywords and the output would be fully optimized listing. Here's an example of a well optimized listing that will serve as our template, right? We are thinking what

we can provide to it to get the same output every single time. So, we want to share some uh some ideas and some examples. So, here I just copy pasted the content Of the listing. I also included uh the images, you know, the the image suggestion should reflect the product and the target customer. And as I said, if I was doing this like for a real business, I would spend more time, more iterations, you know, but the idea here is that you talk to Claude in order to create the skills. No coding, nothing. You're just

having a conversation, right? So I also ask, ask me if you need more details to create the skills. Do Not create the skill until approved. And suddenly he started asking me a bunch of questions. Maybe stuff I wouldn't even think about. You know, keyword handling, intelligently prioritize, do this or that, you know, output structure, do you want just title bullets, image ideas, or do you want backend search terms as well? I forgot about these. So, of course, image suggestions, customization, you know, all of these things he's asking us because we he Wants to create a

skill in a way that we can just get it right the first try to create consistent results. That's what he helps us to do. So, I answered all of the questions and he's like, "Got it. Here's what the skill will do, you know, Amazon listing generator." So, he's now describing what the skill will do and waiting for my approval, you know. So, he's selling uh saying title, how he's going to do that, bullet points, and if I wanted some iterations, I would tell Him. Then the logic, you know, keyword prioritization and so on. And I

was like, yes, let's generate the skill. Then he uses the meta skill, the skill creator, uh, to generate the skill. So as you can see he's creating it and here he's done. So then I would simply download the skill and upload it the exact same way. So as you would see here in my capabilities uh we would have Amazon listing generator generates optimized Amazon listings. And now this Is where really the beautiful part begins because if we go into our skills we have our listing strategist as I mentioned here you know our listing strategies our

project and now the listing strategies if I will work with it in the future he's instructed that he has access to three skills which I will also show you the listing creator the review analyzer and the SQPR analyzer so we can uh essentially have multiple skills within this project in this case We I have three and here you can see a test of the uh we here we can see the test of the listing generator. So what I did I was just hey claude I just added Amazon listing generator skill. Here are my keywords. Boom.

I uploaded an Excel file. If I wanted to make it a little bit more comprehensive the first step would be to go to my Google Drive and get it yourself you know but in this case I just uploaded it and he was like Yes perfect. He opened the skill, understood what I want to do. He extracted the keyword data from Excel and then he created the listing on the first try. No back and forth, no telling him, hey, can you do this? Oh, maybe I forgot. No, everything is already predefined. I just did the thing

and it's there. So, the optimized listing again, he just got a list of keywords, nothing else. and he created a title, he created bullet points, you know, all According to the original template I sent him. He created backend search terms and also the secondary image ideas, you know, just describing what the image should be about, you know, secondary image, lifestyle, hero shot, size and dimension, infographic, uh, basically based on the template. So if you have some brand guidelines like your own framework for creating listings, you would teach cloth and then just upload keywords or maybe

some other inputs You'd like to add. In this case, it was only keywords and search volumes and done. It follows your procedure. You just train it, right? Then we can see A+ content module ideas, you know, brand hero, elevated cocktail experience and so on. And then he also has some notes on the optimization strategy. So, as you can see, no back and forth, no explaining, no long prompt engineering. We've already done all the hard work and now we can just execute the same way Every single time. But we have more skills, right? So, let's take

a look into our skill creator. We also created Amazon review dashboard. So, let's take a look at this one because this one is a little bit more complicated. Here we ask him to create a new comprehensive skill on a high level. We will scrape Amazon listing reviews a large data set. So Python code might be required and then generate a dashboard which will be predefined in our skill. So it also so It's always the same you know it should include summary product sele product section for strength and weaknesses product improvements and listing section ideas for

images to optimize conversion based on feedback. I also um uh added the colors which uh didn't look as good as I expected. So I in the future I would let claude actually to use its own color selection not adding my colors but other than that it turned out pretty well. So it started creating the Architecture of u of the dashboard right because this is more complex. Now we're asking that the skill will be creating a dashboard based on the data set. So we first need to specify you know how many reviews will be there and

so on. So we went back and forth. I'm just answering questions right? Nothing hard. is asking me everything he needs to know. Simply answering questions. Then he creates the dashboard uh creates the file you know uh that I Downloaded and really nothing complicated about it. Just answering his questions and after a few minutes we had a simple skill which I downloaded. And now the beautiful thing about this skill is that it also interacts with an MCP with Ampify MCP. And since I'm talking a lot about MCPS, I have a whole new video uh that will

be linked below talking about setting up your MCPS, right? How to scrape reviews on Amazon and how to do all of these Things right from your cloth. But in this case, it's already included in this particular cloth skill. So, let's go back into our SIS listing strategies that I already ran this analysis before. And uh let's give it a second to load. And if we go right here, we said, "Hey, Claude, I just added the Amazon review dashboard skill. Run an analysis for this as on the Amazon US marketplace." And I just copy pasted the

as and just used which Epify actor to use. If you Don't know what this is, you better watch the other video. But this basically just enables you to download all of the reviews from Amazon through your cloth. So you can then analyze them and get some insights. Right? So he as you can see he ran immediately. Again simple I just copy pasted the asin using the skill and he knows what to do. He downloaded the reviews. He saved the reviews analyzed the reviews and according to the skill he generated the Dashboard. So as you can

see analysis completed what was analyzed key findings but most importantly he created the dashboard and as I said it didn't work out with the colors but the information there is actually pretty good right so as you can see colors not good but let's focus on the uh let's focus on the data we got out of it so just on one prompt give me the insights so he scraped the product you know number of reviews every trading and now the most important thing He analyzed all of the reviews to give us key insights about this product.

He gave us the a breakdown of strengths what customers love and again this is what we ask for. So it will be the same every single time we just scrape another product. The dashboard will be the same. Uh key insights, top strengths, uh and weaknesses, areas for improvement. Right? Since let's say this is our competition, those are some things we could consider targeting in our listing Because these are the things that people are complaining in the reviews, durability concerns, significant number of customers report glasses breaking or shattering easily, some within first use. So as you

can see, critical, common, high impact. And based on this, he also gives us uh product improvements, actionable recommendation, you know, potentially increasing glass wall thickness, add quality control, inlook handling care, uh cart in Packaging and so on. He also adds complexity. How complex would be to implement these changes. Of course, again, I didn't spend as much time providing the context, creating the skill. So, if I was creating it for my own business, I would spend few more hours refining this. So the dashboard always provides all of the information I need, but you get the point,

right? And then he also adds listing optimization ideas essentially how to target the Weaknesses of our competitor. You know, close-up cross-section showing glass thickness at rimle and stem with measurement callouts because this addresses the top concern which is the thickness and fragility of our competition. So as you can see then he tells us to add infographics uh size guide use case potentially some lifestyle images he prioritizes all of these. So as you can see this is a very good breakdown and now it is in a Unified way presented to us no matter what product we

scrape right one prompt great result every single time. But we are still not done yet. Now we are getting into the third one into actual data analysis where Claude is using Python the code to analyze data. So we went to skill creator and this one you actually had to do it in two. We had search query performance right report. Um so if we go up here we just copy pasted this. It's like 1,00 rows and now We're having a conversation with Claude about the dashboard. I don't even know what I want to see in this

case. So I just copy pasted this report and told him like hey what should the dashboard include? So he analyzes the dashboard and he starts telling me you know it should have some executive summary some query performance matrix and so on and so on. So we had a little bit of back and forth designing the dashboard right and then I told him hey my goal is to Create cloth skill but I will do that in a separate chat. I only need your highle instructions about the dashboard so I can copy paste it there. Focus on

the structure of the dashboard. So now he just after the conversation the final dashboard he gave me this. Uh so I just copy pasted what he gave me into my skill creator and I put it there and ask Claude to actually turn this into a skill. Right. So I simply copy pasted the data um the final final result what We want our dashboard to look like and we said you know hey uh create us this dashboard. again. I asked for the colors. It didn't work out. Um, and he created the skill for SQPR dashboards, right?

Because likely you want to analyze these dashboards, uh, analyze the data and it's not always easy to find what's relevant in these reports. And again, this is something we do again and again and again. So, it makes sense for us to have a skill to do it the same Way every single time. Make sure that the way we do it is actually good enough. Because if you take this dash, if you take this data set and just throw it into your AI, sometimes it will be great. It creates first try it's amazing and then five

times in a row it will suck and you will waste time trying to get it right. So now we have the skill and since we created the skill for the listing strategies again we have our SQPR Analysis. So as you can see uh we can hop here and what we did we just uploaded the the data into cloud right we uploaded the the data into cloud and asked for him to use the skill attached report and use the skill as described and it is the same case as with the reviews he creates the dashboard in

the same way every single time right so as we defined the dashboard We get this executive overview, query performance, funnel breakdown. So, as you can see, Key insights, funnel performance versus the market. So, he even compares how our brand is doing compared with the market average, you know, the click-through rates, the card ad rates, and so on. And then he identifies the primary bottlenecks and then we can dig into the data in these sections. As I said many times, like if I was doing this to have like a proper dashboard, I would spend time, you

know, I would refine this. First of all, the colors of course, but Then I would sit down and uh make sure that this is broken down in a way that I actually use this, right? Remove everything I don't need, add things that I'm missing, refine this a few times. But the idea is that once I do this once, then I can reuse it again and again. And of course, if you set this up correctly, you can easily double the productivity of every single person on your team. and and and I can tell you that I

speak to dozens and dozens of Sellers helping them build their teams, their systems. And this is the new way of building teams, right? We don't need an army of VAS. We need a small team of air players that are using AI properly. So, let me show you what that looks like. if you would run a proper audit to build this structure uh of the project, skills, tools, you know, we talked about everything how to actually do this because the main issue is that most people use AI randomly, right? They kind Of have a task and

they just kind of throw it into AI to see what comes out. But we need to start using it intentionally and make it an essential part of our workflows. the same way you do with other software like you don't think when should I use data data dive or helium 10 or whatever it's part of your workflow right you know that you go there in order to get some data right it's part of your workflow and you need to work in the exact same way with AI so What I recommend is to start by running a

quick audit with your team let's say every single team member should identify their top six eight routines you know their tasks and then just pick the three most impactful ones and start building the foundation around it. I like flowcharts, right? I like uh process maps. So I do it this way that we identify the systems, you know, the product development, marketing, whatever that is. Break it down into processes And see what is actually happening. What are the tasks and decisions in every single process? You can do it in a simple way. Just list it in

Excel. Do whatever it takes. You know, write down every single person on your team. Write down the task they are responsible for on a high level. And then you can sit down and prepare the AI co-pilot foundation which if we take a look at this AI or chart let's say you have your employee brand manager. We've been Through this. So what do you need for a brand manager? You need your master prompt. What are the KPIs? Who is this person right? What is their role? Their accountability in the business. Then create projects accordingly. Right? You

need to break this down ideally again by the processes. So if we have listing strategies that helps us optimize listings, then we likely have some product researcher to help us develop new products or find new products, Develop new products. And then each project then has the skills like scale. We had the listing creator, review analyzer, SQPR analyzer and so on. And then every single tool uh skill potentially needs access to a tool. Google spreadsheets, API for scraping, maybe notion, right? Maybe even seller central. So just break it down like this and start building it out,

right? So ideally take a print screen here, remember what the structure looks like And once you have all the context, the skills, the tools, like I I really willing to bet that doubling the productivity, it's really an understatement in this case. Now we are going to build AI automations to put your workflows on autopilot. Data gathering, monitoring, reporting, everything automated. We'll use N8N to build these. You've probably heard about N8N before and if you are like most people, you got intimidated by those Node diagrams, but it's actually way easier than you think. We are going

to learn the foundations of N8N you'll actually use and we'll build a real automation step by step. By the end of this section, you'll have the tools and confidence to start automating your business. So, let's dive in. Today we are going to talk about N8N which is an extremely powerful automation platform. I'm sure you've seen these diagrams somewhere. So now you know that this is Actually N8N which is a notebased automation platform. This means that it allows us to create AI automations and AI agents without really understanding how to code. So first of all we

need to clarify because we will be using this terminology a lot. When we talk about AI automation, we are talking about a fixed sequence of steps, right? Where we are automating a specific process that always go from A to B to C. When we talk about AI agents, we talk about a dynamic Agent that can reason and uses logic. Of course, leveraging the LLMs, the large language models to choose actions and make decisions based on context. So, for example, here we are looking at an AI agent. Now, why does it matter? Why does it matter

to someone like you, to an Amazon seller or a DTOC brand owner? First of all, it allows us to automate operational tasks that eat your time. As I mentioned, these automations can be extremely complex and powerful that will Just take uh stuff off our plate and save our time. It also allows us to connect all of our tools we actually use in business, whether it's your seller center, your spreadsheet, your notion, Slack, all of your data, we can connect it and work with it. whether it's for an automation or for an agent. And of course,

lastly, we can really leverage the large language models and take it to the next level because if we just use a chatbot like clot or chat GPT, we are Basically very reactive, right? We are just prompting the the AI. But here we can implement them into our automations. So they can basically act independently and this is what makes it extremely powerful. Now there are a lot of use cases and once you learn the basics from this video the sky is the limit right suddenly you will see so many opportunities in your business for automations but

to give you some ideas you know you can automate your inventory Monitoring and alerts you can sync data between platforms right taking data from seller central into your spreadsheets you can create automatic reporting you know it can help you with product research competitor monitoring and of course automating stuff like customer support but as I said the biggest unlock here. It's not about the specific automation we are going to create today, even though it's very useful. It's more about showing you the basics and Building a strong foundation and understanding of N8N. Because once you understand how

it works, you understand the possibilities and the next time you do a task in your business, you you work you will realize that a lot of the things you do can be automated because most of the things you do in your Amazon business are done on a recurring basis, right? You are not reinventing the wheel. You are likely repeating the same tasks and processes over and over again. And when you learn this, you will realize that you can actually automate many of them using everything you'll learn today. So first of all, we are going to

set up our N8N account. So you simply go to their website. I will leave all this links in the descriptions. And uh here we basically have two options, right? Either we will host N and in a cloud which means that uh we will pay this monthly fee or you can self-host it which is a little bit more complicated And we are not going to cover that today because the easiest way really is to set up your starter account here start a free trial and follow along to build this automation to actually understand u how how

it works. One little trick, if you are going to, you know, get your subscription, you can ask Chptd or Claude to find you a discount code. You know, I just managed to save like €100 by just asking for a discount code in in Chad GPT. Now, once you set up your Account um in the account, you will see this. We just logged into the account and the first thing we are going to do, uh we are going to switch to a dark theme because I don't want to work with the light theme. Okay. Now uh

there are two main areas that we are interested in. First of all workflows that's this is where we will build our automations and agents and credentials. Credentials are essentially the tools we connect to NAT, Right? So if I check here, you know, there can be my Asana account, my Google Sheets, my Gmail, my Slack, my notion, all of these tools that we want to work with in our automations and agents, we will be connecting uh and then workflows. Of course, we can create folders to keep everything organized and then create our workflows. So I prepared

here one workflow where I want to show you the note types because as I mentioned N8N is a nodebased automation Platform. So we will be working with specific nodes. So let me just quickly go through these. We are not going to cover all of them because there are a lot. There are hundreds if not thousands of nodes in N8N. But if we categorize them into these four categories, it will make it easier to wrap our mind around how this how all of this works. First of all, there are uh trigger nodes. The trigger nodes are

essentially uh a setup that triggers the workflow that triggers The automation, right? It can be manual that it happens every time we click. It can be scheduled you know that it is uh running on a regular basis once every day once every week once every month. We can have a web hook that is essentially triggered through an API call through another application or maybe form submission or when another workflow is executed or by another app. Let's say when we get an email a workflow is triggered. Right? This is Pretty self-explanatory. uh there these are the

key triggers that will be triggering most of our workflows we're going to build. Now one of the most important nodes are the action nodes right these action notes are there to do something in a specific tool or a specific software like your Google Sheets, Gmail, notion, Reddit, YouTube. You know, the key here is that uh N8N has hundreds of pre-made integrations, right? So, as you can see, it can be Send a message in Gmail, create a task in Asana, uh create a new row in a spreadsheet, and these are already pre-made, which makes it easier

to work with the specific tools. So, every time we go here and let's say we search for Asana or any other software, you see all of the actions uh on the right side, you know, actions within a project, creating task, and so on. And this really goes for uh most of the softwares you can think of you know your uh like clickup Notion whatever you can think of but sometimes there is not made uh integration in in nan there's no pre-made integration so what we can do is to use a HTTP request it is essentially

the same thing but we need to define the action through an API call and I will show you everything here in this video it sounds a bit intimidating but actually it's Pretty straightforward, right? The pre-made integrations already took care of the API call and we just choose what we want to do. But with the HTTP request, if there is no action node for that particular software in N8, we can actually create our own action using this node. So we will be using it a lot as well. The next type are the logic nodes. These are

essentially controlling the the path the flow of the data in our workflow, right? So there are if node switches uh merging looping you know uh basically Controlling how the data flows like if you have true or false you know merging data from two sources into one and so on. We will be also using these. Uh there is one important which is a code logic node and this is where we can program and again I know it sounds intimidating but we can program uh how the data will be transformed and I don't know anything about code.

So I will be using something like cloth to actually help me write the code anytime I want to Do something with that particular data set I have and I want to change it into something else. Right? So I will be using cloud as my uh as my development partner here to kind of help me build these and write the code whenever I need it. And lastly an important one which we will also cover today are AI nodes. There are specific nodes called AI agents where we can choose the LLM the large language model for the

chat model. can be your cloth your chat GPT and then We can give it access to specific tools right and this is the difference then between the automation and the agent but the beauty of this is that we can actually connect both of these so we can have partly automation that's taking data processing the data and at the end it's sending it to an agent who decides what to do with it so we can again similarly how we define these uh tools here like asana Gmail we can give our AI agent access to these tools

which is Again extremely powerful. So now we covered uh the node types and there are a few approaches when we develop these workflows. First of all um N8N actually has their own AI where we can ask and describe what we want and N8N will design the workflow for us. This is very useful. It saves time because it kind of builds the foundation. The second uh option which you can use is to ask your chat GPT or cloud to generate a JSON which is essentially a code you can then Upload into N8N to create your workflow.

Right? So we kind of describe uh to claude what we want and then ask him like hey this is an N8N automation I want you to generate the JSON we will download it upload it to create the workflow. But we are not going to do that today because I believe that in order to be able to finalize these automatically generated workflows. You need to be able to understand the basics. You need to build a few Workflows manually because the ones that are automatically generated are not perfect. It's like 70% there and you need to make

make some tweaks to actually make it work which you won't be able to do unless you build something yourself from scratch. So that's what we are going to do today and we are going to use AI as our build partner and in this case I'm going to use cloth. I actually created a project from this with a specific system prompt that he's Designed to be my advisor and I will share this prompt uh down below in the in the description. All of the resources and the template of what we're going to build today you will

find it in the resources. So as you can see this will be my N workflow design assistant. So, he's not doing the work for me, but he will be guiding me through uh through the build. Now, what I want to say is that this video is supposed to show you that you can build these automations With zero experience. Even if you have no past experience whatsoever, this is your first time hearing about N8, you don't know how to code, you don't know anything about automations, you're still good to go. you will be able to learn

the 80% of NAT that you will be actually using to build the uh the basic automations and I also want to make it clear that when we work with clients or when we are building our automations I work with the developer on my team but This time it will be me building it right so you will see that even if I haven't spent hundreds and hundreds of hours building this we will be able to build this live using cloth as our guidance right so this should give you some confidence that you will be able to

do the same. Now, we can finally go ahead and talk about first automation. We're going to set up a competitor price monitoring. It is a simple automation. I believe useful to most sellers, right? It will show us the basics, but at the end of the day, as I mentioned before, it is not about competitor price monitoring. It's about you learning how to use this so you can then unlock so many more possibilities and opportunities in your own business. So we will be creating an automation to essentially scrape competitor listings once a day. Uh here we

have the prompt that we will use with claude to essentially explain what we are trying To do. You know I want to scrape competitor prices daily based on their asent. We will use Appify which is a website scraper and the prices will be tracked in a spreadsheet with history simple date as price columns. If prices change I want to be notified via Slack. It will run daily at 8 8 a.m. So help me design the workflow step by step note by note. Here I have three asets. So we will now get to work. But before

we get to work I want to stress out a few key Lessons here. First of all, expect errors. I will be building this live and some things will break, but it is part of the process, right? This is an iterative process. We go step by step building our automation. But of course, something will go wrong and that's perfectly fine, right? And I want you to see that every time we encounter some uh some hiccup, we will be able to solve it. Next, I want to stress out that it's important to start Simple, really the simplest

automation. That's why we are setting up competitor price monitoring, right? We are not creating this complex dashboard with AI agents on top of it because those things usually break and it's kind of hard to use it, right? The small very purposeful very specific automation say agents will be always more powerful than some one complex all-in-one solution, right? Because we need to keep in mind the compounding effect, right? We need to Understand that if we spend 30 minutes today to save ourself you know 1 hour a week and we do this multiple times identifying little automations

here and there that's where the biggest leverage is right and also I want to stress out document as you build because likely these automations you need to be organized and write things down as you go so for future reference if you go you need to tweak and update you understand what you've actually built because it's Easy you know after a week you build it you kind of lost what was the what was the original logic behind it. So now we will go to cloth and um this custom project and I will copy paste what I

want to do here. Right? So when we when we send him this prompt now he will essentially design the workflow or he will also ask us a few questions how to uh set up the workflow. So as as you can see we maybe didn't even think about a few things we will need but he already Knows so he's asking us how many as we are tracking right because that will affect how we want to store it uh which spreadsheet tool select not notifications you know should it notify at any price change one message per as

or summary of all changes so suddenly there are so many things I even didn't consider and he is asking me to clarify this because uh he then drafts the proposed architecture of the nodes right because he knows nan very Well, and the System prompt is guiding him uh how how to do that, how to explain everything to us. Um and then he explains, you know, alternative if we are tracking less than 10 as which in this example we want. Um he explains the our our options. So what I will do now and we will we

will cut it here. I will just answer these questions and we will get back to it. So as we can see Claude already wrote down the full architecture with all of the note setups which again might seem a bit Overwhelming. I don't recommend to just blindly following it. That's why we are learning this from scratch. Do it manually and just use this as an assistance. Right? It's not that I can just blindly now copy everything and it would work perfectly fine. So what I'm going to do, I'm going to go into my N8N here and

we will create a new workflow. So since we are starting essentially from a blank page uh what I recommend here uh since we said that we are going To track things you know staying organized. So this will be our uh Amazon competitor price tracker workflow. So I will input the name right here maybe even add some tags for future reference. But now I need to begin right and starting from a from a blank page I need to add the first step which is of course a trigger. What triggers this? And eventually we will be adding

uh a daily workflow, daily reminder that will happen automatically. But since we just Start to build this, we want to trigger this manually because we will be testing it a lot. So I want to trigger this workflow manually. And so I will set up this trigger when click execute workflow. And now if we think about it, the first step here is to actually have the asens we want to scrape. Right? So what I'm going to do is I need to create my spreadsheet where I want to track things and also list the asens I want

to track. So let me create a new Spreadsheet. So what we did here we created a very simple spreadsheet Amazon competitive price tracker workflow. You know we have the list of asens here uh which will include and then the tracker we will have date as price. Currently we don't need anything else because we just want to track the uh price in time and the list of asins I included that we want to track three asins. So we will copy them here and now we have the list of asens. We Have our spreadsheet. So the first

step is essentially to get these asens into our workflow so the workflow knows uh which as to scrape and then work with them. So what we're going to do since these asens are in our spreadsheet, we will click click to add a new node and we will search for as you can see Google sheets. Let me zoom in a bit. Uh Google sheets and we need to get rows right because we are our action is that we want to just read the asens that are There. So we will click on get rows and here we

would encounter our first credential setup. I already have my Google account set up here but what you would do is to create new credential and then just sign in with Google. It's as simple as that to get access to your to your spreadsheet. So you click sign in, you click yes and you are good to go. uh since I have uh since I have my Google account here it is you know I will be getting crows Uh so this is my spreadsheet Amazon competitor price tracker workflow and here from the list it's list asens right

so as you can see now I'm editing the note in n8 and what I did I asked that I want to get rows from a spreadsheet I selected my Google account. I said that I want to get the rows, right? Specifying the action that's already pre-made. Chose the document, right? That is my spreadsheet, the Amazon competitor. And the particular sheet is Asense, right? So I basically pointed him in this direction. Now what I'm going to do next is that I want to execute the step to see if N8N is actually able to grab the data.

Let's see what happens. So as you can see now we imported the three asens we have a table here and the three asens are here which is perfect right now I managed to get the data from my spreadsheet into my natan and as you can see it is in section called output And in the middle we have the node and in the left we have input and now as you understand when we are creating an automation that goes a b c so essentially if we are looking at the node b input is what came from

node A and output is what will go into the node C. Right? So that's what goes in. Something happens in the node and then the output goes out. Now when we also take a closer look here, uh you are you're seeing that we are looking at What's called a table. Right? This is a simple overview of the data we just pulled uh in a table. But there are two other ones, right? One is JSON which if you open this it is code right because this is a JSON code that we are now just looking in

a different way at the same data set right so as you can see we still read the asin we still read the row number so it is essentially the same thing just presented in a different way and we will be working with the JSON Code a bit later but it's just a different way to look at it right again no need to get intimidated by the code because we don't really need to understand how the code works. We just see that it's here and we are potentially able to copy it if we needed to let's

say ask cloud for some direction. We can also see it in the table but we also see it in a schema here which allows us to uh see the data in this way where we see the row uh the Row number and the asent. So it's just the same data presented in a different way. Now we executed this and we are happy because we have our asens in um in the workflow right here and before we go before we create the next uh before we create the next node what I like to do and what's

important is that we pin the data because every time we go through the workflow we kind of always want to set up you know the next node working with the data from the previous node. We Don't always want to run the whole workflow again and again every time we test another node. So what we will do is we pin the data so the node will always output current data. Especially useful when we start using something like the scraper or AI and you are essentially paying some money for an action. So you don't want to do

the same action every single time. So we pin it. Right? As you can see it turned purple. The data is pinned for test executions. Beautiful. Now we have um the nodes here. Now we as you can see we have three items you know we clicked and from our row we are getting three items those are our three asens and if we take a look here um one of the next nodes will be actually scraping uh Amazon using epify to scrape Amazon and if you don't know epify you can create your account and actually um on

epify you will get a free account with some credits and these Credits are more than enough for some basic scraping. If you set up your account, you will end here in uh the Appify store. So, you can search for, you know, Amazon. You will see there are plenty of scrapers. What I recommend is to find the pricing model that uh where you essentially pay per event. So, you don't need to pay some monthly subscription or something like that. You're paying it every time um you use it, right? Uh for this purpose, I Actually found one

uh that works very well. It's called Amazon product batch scraper. And why am I looking at this is that I need to understand what do I need to send to this epifi scraper because nent will be sending information to the scraper to trigger the scrape and return me some data right and I need to learn what this particular scraper needs to hear from us essentially uh in order to run the scrape right. So if we take a look at it here, you can see that that You can run this scraper manually, right? So I would

just add my asent here. Anyone can do it easy, right? But I'm using N10. I kind of need to understand what the code looks like in the back end. I need to understand what what it looks like, right? So I can send the right call to send the right data to this scraper. So what I will do is I will take a look at the JSON. That is the code we need to send here. And as you can see, it's simple. It's actually Just a list of asens right here which if I add another asin

from uh my surprise tracker let me add let's say this one just to check uh you know I will add another one here. So this is what if I would be doing manually here but as I said I'm doing it with code and perfect. So it's just a simple uh list of the asens right there is nothing else. If you if I would show you let's say another one you might see that for example this one uh as you can see it's The same when I do it manually but JSON is now more complicated they

want to hear this right so it's kind of different I need to identify what that particular scraper needs to hear from us and this particular one is $40 per month I don't want to pay that so here this one is paper event amazing this is what I need to send him right but as you can see if I go to my N8N what our output it doesn't look the same right if I would send him this he wouldn't know what to Do with it because that's not what he wants to get so I need to

manipulate the data right I need to change it and so it's in a format that the scraper can work with and as you can see we have three items here but he has them batched like this so let's try to aggregate the data as you can see combine a field from many items into list in a single item right so that looks That sounds like something we need. And now when I see it here, when I open the next node, I see The input and that's here. You know, the input that's flowing from the previous

sheet, three items, as everything is here. Perfect. Output, it's nothing yet because we we haven't set it up. Uh, and as you can see, I also can ship, you know, the table, the JSON, everything is here. So, what I want to do is I want to aggregate them. So now we prepared the data for this particular scraper. I want to make sure that the JSON looks like this. And now what we need to do, we Need to fill in the field name. The name of a field in the input items to aggregate together. So for

us, those are the asens. And what is the name of the field is as right? But if we take a look at the scraper, they want us to call it asyn, right? So that is important to keep in mind these little um these little specifics because otherwise it might not work right we want to prevent that. So output field name we want to rename the field and now we will call it A since. So now if we execute the step we test it and as you can see now I'm seeing the JSON right here and

if I compare it with the one that is here right looks the same. So it should work again. I cannot guarantee that it will work but uh we will now pin it and as you can see now from three items we have one item. So now we kind of merged it together and hopefully now it's in a format that in the next step we can send it over to the scraper in a way that he Understands and will send us the data back. Of course we are scraping it to get the price. So I will

go back to my canvas and now I need to call epify, right? And I need to use the API call. If I was going to let's say asana sending an email or something, I could find it easily here. But with EPI, we will do actually the HTTP request which is the API call. We are sending information to EPify and asking for something to return to us, right? We are Triggering this action. So we will go the HTTP request and on the left side we have the our our asensence and now here we need to uh

set up this HTTP request to send the data to our app. So first of all we are not getting something we are actually posting right we are sending the information to uh to epify right we are sending and calling the specific action there is some URL authentication uh headers body sounds quite complicated right but what we are Going to do is we are going to keep it very simple if we have the URL that's essentially the URL to make the request to and if we go here on our um on our app ampify there's a

section with APIs right and if I got stuck here remember I have my cloud here so I can always ask him he already like filled in everything here so he made it a bit overwhelming for us but we can then ask for specific steps it's like hey I'm setting up the epify API help me figure out you know How to make the HTTP request and he would guide you through this uh but here I'm going to go API endpoints that's essentially telling us where to contact it like what is the request to make and here

we actually want to um not only run the scrape we also want to get the data right so we will find run actor and get data set items and as you can see this is the long um to post that is this is the whole URL that's what we need right we are looking for an URL so I will go Here copy it And then I go back here URL paste it. Now we have the URL, right? Um and we need to we are just not sending the something to uh to the ampifier. We we

also want to send them the asense and there's there's s still nothing, right? We we just set up we know where to send it, but what are we sending, right? So here we have what we need to send to the API uh actor To run this scrape. So we will be sending body we will be sending uh the JSON which we created here and the body will be using the the JSON and again the easy way to do this is that when we are sending the JSON I remember I can drag this into here to

kind of specify what I want. So I will go back to expression and I will drag these right here. But actually in this case I want to send the full JSON. Here you can see the result You know and this is not what I want right that that's not what I want to send I actually want to send uh here the asens as you can see in the format so this is kind of the code that's representing what's right here and I want to send the whole thing because I already aggregated here so I want

to send the whole thing I created and this JSON is the representation of what we actually have here as you can see as the three asins are right here and when I Take a look it basically is the same thing Right. So now it should be good to go and when we click execute step we should get the scrape of the data from that actor. So let's see what happens. So we waited for uh like 15 seconds 20 seconds and we got back this. So as you can see it worked. We got some data. If

we got an error again I would go back to my clot and say like hey it's not working. Help me. Uh but thankfully it did work and now we Have this data as you can see. There's a lot of stuff because the scraper doesn't care that we want only lip price. He just scrapes what he's supposed to do. And as you can see, there's usual information what he's actually scraping. Um, so he will return everything, right? 48 as this is what he does. And as you can see, we can also see the runs how much

it cost us to do this. 0.01, right? So basically nothing. So as you Can see, I have five free dollars. I didn't pay for anything and I can uh I can do this easily. So it's beautiful. Now I have let's take a look at the table. It might be easier. So I see the as we have price, currency, quantity, main image URL, product reviews, product images, um and then the raw data, right? And we have this for three items. So as we said, if I scroll, uh let me put it here. Uh we have it

for three items, right? So uh that's perfect. Now let's Spin it because as I said every time we don't want to run the same thing and pay those 9 cents or whatever every single time we we run this. So we will pin the data and now what we what we want to do is we want to track the data in our spreadsheet. Right? So we are tracker. What we want is date, we want as we want price. We don't want all of these things that we scraped into our spreadsheet. Maybe in the future I we

can add some of the things we care About. But in this case, we only care about the date, the uh the ASIN, and the price. So, how are we going to turn this into something that we can actually put into our spreadsheet? So, the easiest way is to actually ask Claude how to turn this data, right? So if I would go here and copy this and now I would go ahead and ask Claude like hey how can I get turn this data into those three data points I need. So let me do that. So as

I said, I asked Claude for help. They I Told him, you know, I copied the copied the whole output I had here from Appify, this whole complicated JSON. And I just simply copy pasted and asked this is the output from Apify. I want you to take this JSON and as it is and write the code based on this JSON to transform it into three data points. Refer solely to the JSON I sent you. I want today's date, as price, nothing else. So he analyzes it and then he gives me this code. I don't know what

it means but I Don't really care as long as it works right. So I will just what I do I copy it and he already tells me node three is the code extract price data. So he said node is code in JavaScript. So if I go here add a new node and he tells me to do code uh run custom JavaScript sounds good code in JavaScript. Nice. As you can see here I have the input right? Looks about right. That's what we are getting from the appy. And here is code. I don't know what it

means, but You know what? I'm going to trust cla and I just copy pasted it right as it is here. I copy pasted it. I don't know what it means, but let's see. Execute step. And perfect. It seems that we got the data we want. So if I look at the table again, we got the date, we got our asens, and we got the price, right? So that's what we wanted. And I can see it again. Schema table JSON. Amazing. Now I can pin the data. And this code node is Essentially one of the ways

to do this. Right? There are other uh other nodes where we could potentially use uh I believe it's called fields or edit fields. Uh edit fields. Right? So we could kind of do this manually that you are manually mapping the the inputs from the previous node. So I I could do that manually but sometimes it's easier to do it with the code if you can ask lot which as I said I created the system Prompt so he writes the code for you even you don't have to understand it as I promised you don't need to

know how to code so so far we got the asens from our spreadsheets we put it together send it to aify and use the code to create the uh the items right here uh so we have three items in the format we need so what we want to do now is we want to go here and actually put them into our spreadsheet. Right? That is as simple as that. So the next step is to track the Prices because in order to compare the prices, we want to track them every day and if there is a

change, so we will be comparing, we want the notification that something happened. So I will go here and now we actually want to write them down. So I'm going back to Google Sheets and I we will be going append row in a sheet. Basically writing into a spreadsheet. So let me find our spreadsheet here. Amazon competitor prices. Here we are going to tracker This time. Uh and now we want to map each column manually. As you can see, we have it beautifully done here. So what is a date? Right? He wants us to map it

out. He he sees date, as price. And he wants us to map it out. So what is date? What is as? What is price? Well, that's easy, right? We will drag it here. This is date. And as you can see, he already shows the result. Uh this is as the last one here this is price. Okay looks good. Execute. So we are executing the the note and as you can see uh he did what we asked him to do here. This the data still the same but the moment of truth is it here? Yes it

is. So we have the dates, we have the prices and we have the asens. So the idea is that tomorrow here we could see uh 25th of October and so on. Uh but what I'm going to do for the comparison because we're not going to wait here for a day to to get another data set. So Actually what I'm going to do is I'm going to copy this and change these days to the 23rd. Let's imagine that this is from data set from yesterday. And let's say one price uh increased the other one will be

like this 35 and the other one stays the same. So we can kind of use it uh as an example uh if if our automation will recognize that the prices changed you know from uh yesterday to today. So one increased one decreased and one stayed The same. So what we want to do now in the automation we wrote down the data. So let me pin it again. And what we want to do essentially is that the automation always goes downloads the data writes it into our spreadsheet. And the next step is what is to compare

the data. Take today's date versus yesterday's to compare it. So now we brought into a spreadsheet. So now we want to go back and get all of the data from our Spreadsheet again. Right? So we will keep it simple. We will get rows in a sheet. We've already done this. We know how to do it. So, I'm getting rows in this spreadsheet in my tracker. And that's pretty much all I need. Execute step. And now I can see something happened like why why is there so many right? What what happened here? We have the 23rd

here 24 the prices everything looks looks good but why is there again this data set again and Again and there's again another page. Why is there 18 items? Well, this answer is actually quite simple because he's executing the workflow based on the number of items and this time we just want to read uh the data. We don't need to do it three times because he had three as so he's reading it three times. Getting rows he does that three times. Therefore, we have 18 points even though we should have six. So, what I need to

do I want to do uh to settings and Execute only once, right? And if I then execute again now I have my six items at 23rd of October 24th my asens my prices right so I will now uh pin the data again right here and my node is ready as I mentioned before um documenting and keeping things organized which is something I told you to do and I'm not doing it here uh keeping the notes you know named properly you know what you're doing here you can change the name here you can add potentially you

know sticky Notes uh as I showed you with the notes. So you can say this section of the automation is doing that you know just keep it organized because the next time you open this you will be like oh what what what did I build here right um so now we have uh this sheet we have six items we are gathering the data from the spreadsheet before the next step I understand this you know might not be the best way to do it but the way I build it is just so it works right if

You spend a little bit more time look at the more you know more optimized way to do this you might figure out something better but here we are really trying to stick to the basics and show you how to do it. So if you know how to automate and don't hate me for the for these steps that you would do it better. Of course there are better ways to do it but we're just trying to keep it as simple as possible. So now we have this output and what I essentially want to do Is that

I want to compare today's price with yesterday's right and I want to do that every day. So I can say that if today price changes I know that. So I'm the only thing I'm doing is comparing today with yesterday price for each and every asin. But since we are tracking the prices every single day, we kind of need to specify what is today and what is yesterday, right? And again I will do this with the code note because why why would I try to figure this out when I Can just go to Claude and tell

him what I want to do. So I will again tell him like hey I have these this is the JSON code I have and help me do the thing I want to do. So now we did the same thing we copy pasted the JSON we got that the output you know this is my output JSON write me a code that will do the following. It will compare today's prices with yesterday's and identify whether there was a price change or not refer solely to the JSON I sent you. So He sent me this uh but actually

what I wanted to do what I noticed one thing uh I asked him for a minor change because he merged the data. So I just ask him you know I need to compare each item separately. The output should include true or false whether there was a price change and then by how much. So he writes the code and as you can see the code is getting a bit complicated but it doesn't matter because with cloth it usually works. So and if it doesn't we Can ask for more guidance. So what I can do is I

go code here I have the input uh I copy paste the code here execute the step to see what comes out. [clears throat] And uh here again we have our table. So we have our as price change true or false. That's perfect. That's what we need. Uh yesterday's price, today's price and then the difference. So now we compare the prices from today and yesterday. We have for our asins. Perfect. So we are getting near the end, right? We are keeping it uh on time. So let's take a look. The next one will be actually figure

out which of these changed and which didn't because as we said we want the notification on slack if something changed notify us on slack and since we are saying if we will find the if note um and find basically comparing whether it's true or false so uh is true or false right so true means price change so I will go true Uh if price change is true. And as you can see, we have true branch, two items, price change to true true, false branch, false. This one didn't change. So for the uh for the false

branch, as you can see, one item going into the false one, two items going into the true one. For the false one, we are finished. We don't care if the price doesn't change. We don't need a notification. But for this one, we want To be notified on Slack, but we just don't want, you know, uh to see just, you know, yesterday's price. We want we want something a little bit nicer that is formatted as a notification for Slack. So what I'm going to do actually is I'm going to uh take this the JSON and now

again I'm going to use the code uh as clot to actually hey use the data to summarize it into one Slack notification and use emojis make it nice you know and and comment on that. So we Have the code. So he transformed this into a nice message on Slack. We can easily read, you know, we can then post it into our price changes notification channel. Uh so let's copy this and do the thing. So I simply ask lot posting the current JSON I have. Now I have this JSON. Write me a code that will prepare

as a message for Slack refer solely to the JSON. The notification will all of the ice in one message and include some nice emojis for easy readability. So as You can see, he prepared it here. I like it. So I'm going to copy it again. Go here. Find my code node code in JavaScript. Uh copy paste it here. Execute step to see what's going on. And here we have the message. Looks good. You know message price alert price in yesterday today increase by one. Amazing. So we will pin it. And now the last step. We

have the message. So of course the last step is Slack. What we will be doing. As you can see a lot of Actions here. We will be sending a message and so we will send message operation send message to. So um we will send it to user or channel. I will send to user uh and from a list again I will find myself. So I will send it to myself. And now the message text. What is the message text? Well, we can again drag this message as an expression of the JSON because it will be

changing, right? That is why we are using the code for That because it's changing. It's a variable. Uh so we need to include it as the JSON. And now uh the moment of truth, we will click execute step. We see something happened here. JSON uh but we don't care about that. What we care about is our Slack. And as you can see what came to me here. Amazon price alert as yesterday 3435 increased by $1. This asin decreased by two. Two price change detected automated with N8N workflow. So now you understand how to Create a

simple automation. We learned quite a lot in this uh in this simple one because now we can combine it. We can send it. You can send it via email. You know, you can connect more action items, what to do, right? You can gather more information than just prices, right? Really, the sky is the limit. Now you understand the basics. Now you understand how all of this works. And hopefully you are not as intimidated because maybe I made some mistakes. Maybe it could be better. But the most important thing is it works. You know, maybe if

I had 200 days since it would probably break and I would have to explain to my cloth like, hey, actually I need to batch it and do a little these advanced things. But for now, this automation works very well. So, what I'm going to do now, we're going to pause the video and I'm going to add new asins and just run the automation u again to see, you know, what happens and if it All works out. So now what we did, we added a new as into our workflow. We also added some data set, you

know, so we have like a two uh 22nd of October 23rd. So now we have a little bit more data here. We have another as and what we're going to do now is to run the full workflow uh once again. So I will execute the workflow and as you can see you see how it's working through the nodes. This one is taking a little bit longer because that is the epify node You know that kind of needs to scrape return the data he already did. Now it's running uh and it's done. Workflow successful. And as

you can see I received a new notification here. Amazon price alert. I see the asin yesterday today. yesterday today increase decrease increase decrease so the automation is working very well and we are done here so we can save then I would click you know replace the trigger here with the schedule so what I would do you know Schedule trigger I would do days between triggers uh start every AM and then remove this one connect it here you know like this and then activate the workflow and it would run every day at uh 8:00 a.m. So

that's pretty much it for this automation. But now I want you to think about one important thing. We didn't really use AI in this particular automation, right? We built this. But what this allows us and what I will cover in the future videos is that as I Mentioned, if I go here, yes, I want to save this and I create a new workflow. uh you can find that there is actually an AI agent and I want to show this to you because I want to cover this in the future videos because here uh the AI

agent actually is where we implement the AI into the workflow the large language model and as I mentioned before here we choose the chat model we choose the memory and the tools he can interact with right so he here you can choose Your open AI you know you can choose uh cloth in memory. You can usually use simple memory here. Uh but more importantly, you can give him tools. You can give him tools like hey you can now work with Asana. You can create tasks in Asana. You can do X Y and Z. So what

I could do at the end of the uh automation instead of sending me a notification to Slack, I can actually send it to an agent and tell him to do more with it. Because uh what I want to do here, I Also want to show you just a quick example. If you create like a very simple AI agent just for the purpose of this video, uh we created an AI agent that is connected to a model. Uh he has access to spreadsheets and he has access ability to create tasks in a sauna, right? It's very

straightforward, very easy. Uh and here we have a chat. We can actually um we can actually talk to him, right? The same way you would talk to your large language model. So we can Talk to him right here. And these agents have also a system prompt, right? So essentially you are you you tell him what he's supposed to do with the data he is getting. Right? So this is just a simple prompt that he uh this particular one can is a meeting assistant and he processes meeting transcripts and creates actionable tasks. Right? So if you

are interested I can cover these in my upcoming videos because what this can do it's really extremely powerful Because uh in this example I just put like a meeting transcript right here into into Google sheet it can be wherever but in for the purpose I put it into Google sheet we created a sana workspace right here right so in a sana workspace uh for him that he has access those are the tools he has right Google sheets and asana as you can see the same thing we did just right now you know transcript sheet one,

he has access. He knows everything what to do with it. And Now I can talk to him. And this could be triggered automatically or via chat. But I can tell you like, hey, can you use the transcript from my spreadsheet, identify the action items and create tasks in Asana accordingly. So we send him a message and he see we see him working now. He went to here. He took a transcript. He's thinking he's thinking identifying the uh identifying the action items for the transcript. So You can see how he's operating. This is actually quite cool,

right? That you can uh you can see all of this. And now he's creating task in Asana. He says that he created six tasks. Now he's waiting creating a reply. And now he told us I created the following task in Asana based on the transcript. He created the due dates uh based on this transcript, you know, the tasks. So, as you can see, he described the Tasks with the uh with the due dates in a sauna. And this is really just the simplest possible thing. Now, imagine combine it with the automation, use the agent, give

him the tools, and as I promised, you've learned the 80% of N8 you are going to use if you want to build your automation. Now we are going to create professional product photos without spending thousands on photo shoots. When we scaled Elixir Glass 4 to 8 figures, we Spent tens of thousands on photographers, studios, models. But today, you don't have to do any of that. We'll use Google's Nano Banana Pro to generate images that actually convert. We'll build a complete listing from scratch so you can see the full process. And by the end of this

section, you'll know how to create listing images, lifestyle shots, and beautiful infographics. So, let's dive in. Now, I'm sure you've heard about Nano Banana Pro before, but just to clarify, this is the image generation model that comes with Gemini 3, which is the AI from Google, which means that we can get access to this directly from our Gemini account. So, we will simply go to Gemini from Google and we will get access to Nano Banana Pro right here. So, we will select create image and just make sure that you always select the thinking model which

chooses the Gemini 3 Pro. Now we can also get access to this image Generation model from thirdparty software like Hicksfield that basically brings all of these models from OpenAI from Google under one roof paying one subscription. But for our video right here we will be using only Gemini because it is more than enough. When we generate images we can use two approaches. We can either use a detailed description of the scene or we can combine multiple images. So let me show you. In this case, we have a product Photo. In this case, it is a

snowboard. And we can use a very detailed scene description in order to generate this image. So as you can see, there's a lot of detail, technical specifications, and I will show you how I generate these later on in this video. So then the Nano Banana actually creates a photo like this, which is exactly based on the description. And of course, Nano Banana works very well with text as well. So we can add uh a text overlay into our Images. Right? In this case the text is not very nice. So we simply ask to change it

so it fits more the branding of the extreme sports. So it creates something like this. The second way to create images is actually to combine multiple images. So what we can do is we can give a brief description. You know a couple enjoying a romantic dinner. We include the photo of the restaurant of the woman and in this case let's say the product is a hot sauce and of course the Photo of the man as well and just say you know create this image using the hot sauce closeup 1:1 ratio and we get the image

like this but as I said um Nano Banana Pro is actually extremely powerful so we can not only create images add text into them we can also make minor changes in the images without changing the full image right because rarely ever you get the best image on the first try you want to make some minor adjustments. And this is where it Becomes extremely powerful compared to the previous models because we can say like, hey, let's let's show that he actually added way more sauce into his dish right here. And as you can see, he simply

changes the photo and adds more sauce without changing the rest of the photo. And this is beautiful, right? So, because we can then change the image to really match our vision for that particular photo. Always uh download the image if you want to see the full Quality, right? And I forgot to mention this because if you see the photo and here this is just a preview. So you may see that the text on the bottom might not be exactly accurate but if you download the photo it finalizes the rendering and the quality of the photo

will be higher. So before you judge anything make sure you actually click the button here to download the full size of the photo. Now the only thing to stay away from is a signal conflict. Right? This is a simple explanation here because the model is fairly easy to use, right? As you can see, fairly straightforward. We just say what we want. But if we combine these two approaches, right? We show AI the images we want to use. But then we have the signal conflict, which means that we actually start describing what we already shared in

the photo, right? Because here we share the photo of the restaurant and you can see in our photo It is exactly the restaurant. But if we would start describing the restaurant as well, you know, this a cozy Mexican restaurant uh with green door and so on, the model will get confused and the results will be horrible. Right? So for example, here is one image of the signal conflict. Right? We were talking about the view. We added also a picture of of a Greek island of the view, but we also describe it and now the model

is confused. So make sure you avoid the Signal conflict. If you are using one of these approaches, make sure you stick to it. either a lot of detail in text and just the product or a combin combination of images. Here I believe it goes without saying of course make sure that when you're using images like this that you actually have the rights to use the image in your listings. Now we are not going to create images just for fun. We want to create images for our listings to optimize our clickthrough rates and Our conversion rates

which means that we need a good strategy. So in this video we are going to work on this listing. we will redo a whole listing of this brand of olive oil because if we take a look at their listing, we see that their images are not the best. Even though the brand looks beautiful, the images are a bit weak. As you can see, wall of text, not very no low lifestyle images and overall the photos are not very nice. So, there's a lot of potential to use AI To optimize this listings. And to be honest,

most most listings on Amazon still look like this. Maybe even worse than this, even though you can really fix this with AI in just a few minutes. So in order to create these images, we will need a strategy. And I like to use Claude for this. So when we go to Claude, we created a project as a listing strategist. We taught this particular AI to be our listing strategist, right? To create a strategy For a listing in order to optimize conversion rates and clickthrough rate. So, we don't want beautiful images. We actually want images that

drive sales. So, what we did here in the knowledge files, I summarized all of the branding of this uh of this product. I gathered all of the text from all of their images from their website. So, we have all of the details about the product, right? So, we don't want to hallucinate in the images. We actually want to sell share Real data about this brand. This would be much easier if it was your brand, but as now I'm just scraping their listings. But more importantly, what we did, one of the most important steps that pretty

much no one does is actually scraping their reviews. So what we did is that we downloaded hundreds of their reviews in order to analyze them. So we can see why people buy their product, but we can also see what people don't like about the product. So we can potentially use These insights to create better images. So as I said, we are not just creating beautiful images. We actually need a strategy, right? So, I'm not going to cover this in this video, how to actually scrape the reviews, but we have these data sets. So, we have

hundreds of reviews from their listings, five stars, four stars, and so on. And everything is right here, which means that we would uh then feed it into Claude and Claude will help us generate the insights from all Of their branding and from all the reviews we scraped. And since now we have all of the data in cloth, we can ask him to begin our analysis with our listing strategies, right? So first we want him to write all of the key insights about the brand. So as you can see, product DNA, uh, Fresco Baldi family, premium

positioning, distinct flavor, cold press within hours. So we have the information we need for our listing images, but then we also ask him To analyze the reviews. So he opened all of the files and analyze all of the reviews we had. So we had 384 reviews from their listing. As you can see here are the ratings. But what we care the most about if there are any insights we could then use in our images. So first of all the one star complaints. Some of them are complaining about bitter peppery taste. Right? But since we know

that this is the unique taste of this particular olive oil, we might consider Adding this into our listing. Right? to kind of warn them or rather inform them that this is how it's supposed to taste, right? Price value, right? So, it means that we need premium positioning and uh then we have fivestar praise, people talking about flavor, quality, authenticity and heritage, right? This is the real deal and so on. So, we also want to highlight that because this is the reason why people buy it. So, we definitely want to showcase that health Benefits potentially gifting.

And now this listing strategies actually tells us the implications for images, right? Considering everything we've learned so far, we must address this pre-educate them on the peppery taste. You know that is the that is quality indicator not a defect. Then we need freshness, reassurance, price, the justification and setting proper expectation. Then this is authentic Italian not supermarket mild. Now uh we ask him to Give us the sequence of the images right the recommendations and as I said this listing strategies is trained to give us specific ideas for images. So, as you can see, we get

a few ideas, but I'm not going to waste your time here on this video going through this into too much detail because it's not about the product, right? It's about the approach. And then I share my ideas, right? It's kind of back and forth clarifying the strategy. So, I'm saying like, hey, for Image two, I would like some emotional hook like something like a taste of Tuscany, you know? Uh, for image three, we should have a close-up of a beautiful meal with the olive oil. Image four, that it's only the finest olives. And image five,

the bold and peppery. you know this is the authentic Italian right so this is important to define the strategy and now we're just asking the AI if uh we covered all of the important points right considering the reviews and Considering everything we know about the brand so pepper taste education freshness reassurance price justification heritage positioning and usage scenarios and after a little bit of back and forth we asked the AI to give us the brief outline of all of the images we agreed on so as you can see seven image sequence right here so We

will put it uh into our board right here. And if we take a look at it, of course, we have the main image, heritage Lifestyle opener. This is the emotional hook. Image three, primary use case. Image four, dramatic bottle half submerged in olives, dark moody lightning. Image five is going to address the bold and peppery taste. Image six, bottle surrounded by Italian dishes, the heart of every Italian meal. So this is the draft. Now we start generating the images, the good stuff, right? And the first step I noticed is if we go to their listing,

I noticed That the bottle is actually quite dark. And then I checked the reviews and the bottle looked something like this. So as you can see, you can see the olive oil. It's it's a transparent glass. So what I did, I already started generating the uh the main image. So let's go all the way up. And I just ask him, give me a nice shot of this bottle on the white background, crisp colors, good softbox light. Put a few olives at the bottom of the bottle from each side. So, as you Can see, this is

it. Looks 10 times better than the first one. But we are not done yet. We will also ask him to add a nice hang tag on the top that says made in Italy that would fit the branding. And as you can see, we get something like this. Come on. That's was the second uh image we generated right here. And it is amazing, right? This is this is a perfect photo. So, just for comparison, let's take a look at this one and then this one, right? So, I'm Pretty confident that this one would be the one

getting more clicks. So, our main image, that's pretty straightforward, right? There is not much thinking through it. We just created a beautiful image that represents the brand. As you can see, all of the branding on the bottle is pretty much perfect, right? It is possible that there will be some minor tweaks we will have to do in some of these images. Um, but that's easy, Right? You can pay someone on Fiverr to do that for you to kind of fix the letters or something like that. But 90% of the time it really it really fits.

So let's begin with the first photo. Right, we have the heritage lifestyle opener. So what we're going to do now is we're going to go into Claude to tell us the specific prompt to use with Gemini. So let me do that. So in our project, we asked Claude to write the instructions for image 2, the second image we agreed On, and he's trained to write the scene description, potential text overlays, and also technical specifications. Right? It maybe doesn't have to be as detailed as this, but I got the best results when I actually try to

be as specific as possible. So now what we're going to do is we're copy all of this just to make sure that text overlays a taste of Tuskanyany uh bottom third over 10,000 years of Fresco Baldi family traditions and a small circular batch Fresco Baldi estate. Right. So now let's go to uh Gemini, copy paste these instructions and of course we also want to upload uh the bottle the product itself. And now we will ask him to generate this image. Okay, our image is done. Uh just to be sure, I usually ask to redo the

picture. So I get like two variations. So as you can see, a taste of Tuskanyany over 10,000 years. Here we see our product. And I personally like this other one more. As you can see, it Kind of goes with our vibe, the Tuscanany, uh the wine, the meals. I think it looks perfect. So of course usually I would download the image go here and our heritage lifestyle opener is finished right. So as you can see the branding it all fits very nicely and now I would go to Claude and ask him for instructions for the

next image. The next image is primary use case. We got these instructions right here. We copied them into Gemini and we got our result Right here. So as before I click to reduce. So, I have two variations and I personally like this one more. So, let's take this one, put it into our board right here. As you can see, the the branding really works here. Right. So, ideally, I would also put this into Canva or Photoshop to kind of tweak it, maybe the colors, make sure that it is all aligned. But to me, it looks

pretty good. So, we can continue with the next image. Image four is the dramatic bottle In olives. So, we copied our instructions and it is generated uh in Gemini. Right here, we have two variations, but in the second variation, you may notice that he actually changed the picture on the bottle. Like the text is fine, but he somehow changed it. So, we're going to use this one. And I personally like these close-ups a lot because we can actually see how well it handled the text. And in this case, um there will be some minor issues

here and There, but again, it's fairly simple to fix this. Plus, as I said, we will now download the photo and see if it looks better once it's upscaled. So, as you can see, this is the image we downloaded, and the quality is just incredible. And even if we zoom in on this, that is actually what's on the original bottle. So, as I told you before, if you see here, the quality is not quite right. You you see that there are some AI artifacts, but when we take A look at this, this is the full

scale of the photo, right? The detail is incredible. Like you couldn't really tell that this is AI generated. Really, this is uh this is honestly quite insane. But we now copied it back. Here we have our dramatic bottle in olives. And our next image is the extreme close-up of the spoon. So let's get the instructions and generate it. This one is a simple image, but I believe we are addressing the issue that some people Were complaining that it's too bold, too peppery. So we say this is the mark of true extra virgin quality. Right. So

again, the photo is amazing. We're showing the liquid, right? If I had the product, I would actually check whether the color of the liquid is accurate, but I believe this looks good to me. So, I will put it right here into our sequence. And again, we can generate the instructions for the next one and copy it into Gemini. Since we did the Groundwork and we prepared the project in cloth, now we can generate the instructions very quickly, which means that we can also generate the images quickly. So, now we have the bottle surrounded by Italian

dishes. So, as you can see, the heart of every Italian meal. It looks good to me. Perfect for salads, vegetables, pasta, fish, and bread. So, I'm going to copy it right here. And as you can see, in like 10 15 minutes, we are almost done. We have one More picture to go. And the last one is with badges. But of course, graphic design is also not a problem for Gemini. So, we took these instructions and this is the result. Again, we generated two versions. And as you can see, the designs work pretty well, the colors,

the product. So I personally like this one more. The composition looks great. And we will put it right here. So the final step here would be to review the images and potentially go back and tweak It so it it basically fits our vision and our brand and all of the colors are fine. We can do minor tweaks in Canva or Photoshop. But if I would rate this, the main image for me is 10 out of 10. Looks amazing. The the lifestyle emotional hook again perfect 10 out of 10. This could u use some work, right?

It feels a little bit empty. Maybe we could see more of the product. Of course, the label here was a bit wrong. So, we could redo that. And again, as I told you, it Fixes minor uh it can do minor changes in the image quite easily. So, you could tell it like, hey, make sure the logo on the ball is correct and very likely you would be able to fix it. I like this photo here, but I personally maybe would go for a lighter photo because all of them are very warm, very light, and this

one is maybe too dark, too dramatic. So, we might change it. This one to me is perfect. We're addressing the issues from the reviews and hopefully kind of Uh educating people that this is how it's supposed to taste, right? There's nothing wrong with it. Here, the colors are a little bit off. So, I would then say, you know, this banner color should be this, you know, a little bit creamier color, not white. But overall the image is beautiful. And again this minor change here that's uh that's done in a few minutes. And again lastly this

one looks beautiful. Right? We can see the product. Maybe we would put the box also There. Right? So I have the image of the box. So I would maybe put it uh behind it. And then we have all of our images. I also spent some time you know playing with it a bit more because for the purpose of this video I want to show you how relatively straightforward it is if you do the groundwork. But you can play with it like you can redo the images, play with the ideas. And here are a few we

generated before. You know, Aramas of Tuscanany, as you can see, this is Beautiful, very appetizing. The heart of every Italian meal. I did a few variations. And I personally like this one more because it really looks centered. You know, the heart of every Italian meal. I think this one is a bit better. Then we tried a few more lifestyle. So, as you can see, we have also the apron, you know, as a chef. this one like worthy of social media, beautiful potentially uh creating even multiple images in one like this, right? So, we just ask,

you know, we want framed four photos like this and it can put it all together in one try. You know, all of the use cases. For example, this I also liked very dramatic but a bit warmer than the previous one. These were some variations on the taste of Tuscanany as you can see the cutout behind her head. So it it really is beautiful and like good luck telling that this is actually AI, right? So we have the finest from Tuskan Hills and This could potentially be a banner for our A+ content. So as you can

see, extremely powerful. There's literally no excuse for people to have bad listings yet. I believe that there is this window maybe 6 to 12 months before everyone can click a button and recreate their whole listings. Which brings me to my last point. You can actually create scalable automations for your products. either you have hundreds of products and recreating listing, spending 30 minutes, 1 hour, two hours per listing is just not feasible. So you can automate this process. Everything I did having the agent writing the instructions, feeding it into Gemini, you can use an 8N to

create this automation, you know, to just upload a photo of the product and everything is automatic. Or you can use a scalable automation for social media like let's say that this brand would create content on Instagram and we can create these let's say carousels for Let's say one of their strategies is showing recipes right so this would be a recipe for spaghetti using the olive oil and it can break down the steps like this so first we show all of the ingredients step one all of the text all of this is done with AI right

beautiful carusal look at the colors this would really pop out and you can really generate as many as you want. You know, it's extremely extremely scalable. Potentially, you can also try creating Videos with the uh with the other model right here. Now, we are going to automate image generation. Instead of manually creating images one by one, we'll build a system that generates entire listings with one click. We'll use N8N combined with Nano Banana Pro to create an automation that takes your product details and outputs professional listings that convert. By the end of this section, you'll

have a complete automated image generation System. So, let's build it. In my previous video, we took this average listing and used Nano Banana Pro inside Gemini. And after a little back and forth and a few tweaks, we ended up with listing images like these. So, as you can see, for me, these are perfect. But today, I want to take it a step further. I want to generate these amazing listing images instantly with one click. And for that purpose, we created an AI automation that works like this. So, let Me show you. We have a product

here. And as you can see, it is a jar of honey. And all I have to do is to take the AS and fill in a little form. All I need to fill in is the AS. So, let me fill that in. And then I simply click submit. And that's all I have to do. Now, we will wait roughly 5 minutes before the images are generated. What's happening in the background is that the AI automation is actually scraping the listing on Amazon. It's looking at the images. It's Understanding what the product is all about and

not just the product but also the reviews to understand the language the customers use, the issues they have, why they buy the product, why they don't like it. So the listing images can really tackle these particular points from the reviews and from the specific listing. So let's wait a few minutes and I will show you the result. So our images are ready and first of all we received an email which kind of Summarizes the strategy for this particular product after scraping it and fully understanding it. So the AI agent determined the strategy as you can

see the brand aesthetics and then description for each and every single image as you can see emotional hook uh image three primary use case and so on. But what's more interesting are the specific product photos. So let's take a look at it. Let me switch in here. And these are the images we Generated automatically by just adding the as. So let's take a look at it. Number one, we have the emotional hook. As you can see, the image looks amazing. And we have our product right here. Let's take a look at the next one. Right?

This would be the primary use case. Again, very beautiful. And now here, this image is like what makes the product special, you know, handpicked from beponsible keepers. Be responsible. I actually looked it up and they use This as their own trademarked word. And here we actually have the objection reframe because if we take a look at their listing, some of the reviews are actually mentioning that the customers are not very happy how the honey crystallized. So now this is actually tackling the objections because this language used in these images is based on of what the

customers used. It's using the words why they love the product, but it's also tackling the Objections and the issues they might have before uh before buying it. So, as you can imagine, someone is reading the reviews and the most negative reviews are that it's crystallized, but we have an image here explaining that it's not an issue, it's actually a feature of this highquality product. Then we have your kitchen essentials. You know, this is kind of the versatility use for the product. And lastly, we have uh summarizing what's special about this Product. premium raw American honey,

US grade, true source certified. As you can see, all we had to do put the AS in and all of these images were generated. Now, I will show you how exactly the workflow works and as I said, we build it in an A10 and if you want, you can make a copy of this exact workflow. I will leave the link in the description so you can set it up and use yourself in your own business. So, as I mentioned, all we need is the input. Uh we actually have Uh two versions of this. We have

a manual input like if you have a new product and there is not an existing listing yet you can summarize all of the information that you want to include there and then we have the automatic which I just showed you right we use the asin put it in there and uh the listing is generated so higher level what happens in the background is that we have the data gathering we're using aifi to scrape the listing to understand the Title the bullet points download the images right but also analyze the reviews where AI agent actually looks

at the reviews and understands what's in there uh I will be also summarizing the cost so you can see that this is very very very affordable compared to regular photo shoots or working with graphic designers because the listing scrape is around 1 cent per listing. That's pretty good. Uh then we have the listing strategy. So once we have the data, it Sends it to an agent that uses the reasoning and is prompted to actually be a listing strategist to design the strategy for the listing, you know, to design the sequence of the images and so

on. This is around 7 cents per listing. Also pretty good. Then we generate the image prompts. That's what we've seen uh right here. You know, these are very necessary sending to Nano Banana Pro. So they can actually generate these images, right? Those are very useful. So we have To generate the prompts and then send it to Nano Banana Pro to generate these images. This is the most expensive steps is around 40 cents per image depending on the resolution, but in this ca case it's around 2,000 to 2,000 pixels around 40 cents per image. Still very

affordable. And then the delivery. So, we received the email, we received the images on our Google Drive and so the total is roughly $25 per listing. So, it's very affordable and very useful. So, we can use it at scale. You know, if you have 100, 200, 500 SKUs, still very doable, definitely worth it. Uh, so let's take a look at the workflow. As I said, we are using N8N. And if you've never used N8N before, you might be quite intimidated, but don't worry. First of all, I have a detailed video on my channel which I

will also link in the description where I explain all of the basics of N8N. N8N is an automation platform. It is a notebased automation Platform. So it is a visual builder something like this and I will show you the specific workflow in there and explain everything how it works on a high level but it would definitely help you to to understand the basics and also it's incredibly helpful, incredibly powerful. So even if you don't want to use specifically for this, this will allow you to automate pretty much anything because you can combine AI, you can

create well basically pretty much Anything. And if you want to use this specific workflow, as I said, you can just copy it. I also will include a setup guide. So if you don't want to learn anything from NA10, you can still use the setup guide just to follow the steps to activate it in your own business because as you can imagine, we are using the API keys. So you just uh to access these AI uh tools. So you will need to replace those to use your own. But other than that, the setup is fairly Straightforward

after you make the copy. So let's head into N8N. And as I mentioned, we have two workflows. One is the manual where we essentially fill in the information about the product. Uh as I mentioned, uh when you don't have the ASI, maybe it's a new product. So you input product name, product details, print identity, and then you upload the product image. And it's pretty much the same thing, right? So I'm going to focus on the automated workflow since that's Going to be relevant for most people, but it really changes only the stage of the input.

So let's go into into the workflow. Let me zoom out a little bit. And this is the full workflow, right, with all of these explanations. And now we're going to go step by step. So the first thing I'm going to do is actually I'm going to uh upload the the last run we did. Uh as you can see, it took around 6 minutes to generate all of the all of the images for the honey. I will Copy it into the editor so I can show you the specifics. Right? So let's take a look at the

beginning. Again I understand might be quite intimidating but don't worry I will explain everything. So at the beginning we have the form submission. That's where we put the as right that's all there is. So once we put the as in it goes into the scrapers and as I said we are scraping the listing itself the images and the bullet points the title. So that is the Listing scraper right here. It is using Appify and Epify is a website that allows you to scrape websites like Amazon, right? Um and here we can see once uh this

is sent what the what N8 gets back from EPify. So as you can see on the right side uh we have information what we received after we asked Empify to scrape this particular as right. So we sent this as told hey we want information from about this and we received this. So, as you can see, we Have the URL, we have the title, we have all of the features, we have the image listing, uh the image links and product description. So, anything you might want to know about this product, everything that's on the page is

now here. Also on simultaneously, what happens is a review scraper. It's the same thing, but this time we are actually scraping reviews, specific reviews from customers, right? So it basically looks up the as looks at the reviews and start downloading the Specific reviews and what the customer actually said about the product. And as you can see text my grandson loved the honey, right? So as you can see we we had uh multiple reviews right here. What happens next is that uh we basically need to change the data that returns so the next agent can work

with it. What happens with the image is actually we need to get the main image and we also download one secondary image because that's what's going to be sent to Gemini So it understands what the product looks like right because as you can see here in the images it looks exactly as the product so it uses the main image for that and uh once it downloads the image you can see view this is what we see what it downloaded right again it used the link that we scraped before so it downloaded the image and in

order to send it to the AI I again I don't want to get too technical but in order to send it to the AI we need to change it To essentially a text string right because we cannot send with API call just the file we basically need to change it into something that the AI understands right you don't need to understand what this actually means but it's essentially turning the image into text that we can then send to Gemini so it understands what the product looks like right and we do this again in N8N so

we are extracting from from the file and essentially changing the format of The file. What happens simultaneously and this is the more important part. We take the product information from the product listing you know the name and bullet points because we want to send it to our agent the listing strategist who analyzes uh the reviews and everything about the product. So first of all it aggregates all of the reviews we gathered and it merges the information from the product listing. So essentially what we are left with is the product Listing data uh and the product

description and all the information from the reviews. We're kind of merging everything together. So we can then send it to the listing strategies. And this is the main part where we have AI playing a big role here. Right before we generate the images, we need to design the strategy. That's why we have the prompt generator right here. And the prompt generator actually has a very detailed uh prompt That it basically understands uh how to design these uh listings, how to design these prompts, right? So if we just take a peek on how this AI agent

is trained, we have our system prompt right here. So let me open it so you can see it. So it basically says you are an Amazon listing strategy specializing in secondary image optimization and so on. And then there's all of this design philosophy, you know, describing how the images should be structured because as I said, we have The emotional hook, right? That's the best thing for most products to use as image too, primary use case and so on. So this this prompt really includes pretty much everything to create these beautiful images, not to mention the

specifics, how to describe the image for Nanobanana. So it really knows, you know, how to use the light, how where to use a closeup and all of these things. So we spend a lot of time actually designing this system prompt to really Get output like this. So now the agent receives all of the information we scraped about the product and the reviews and it creates um the output that's necessary for Gemini for Nano Banana Pro that will generate it. I will also show you uh what it looks like if we use a different model because

for this agent we are using claude to create these listing prompts right what I showed you the ideation behind it the description of the scene we are using Cloud but for some of these we also use Gemini so you can see how these two models think differently the system prompt will be identical but some of the images then I will show you generated by Gemini and some of them I bar claw so you kind of see how these two models think differently Okay. And once this uh designs the prompts, all of this is merged, right?

The images we had here and the prompt we generated. Now it's split into specific prompts Because we are generating six separate images, right? So it basically takes all of the data um merges the information about data one, you know, that's image one, data two, it's image two. And then the output, you know, image number, image type, emotional hook, colors, overall feel, scene description. So all of that is right here. Now we just split it out. So each and every single prompt is separated. And now we are going to send it into a subworkflow, which is

Essentially a workflow that sends these to nanobanana. And this is basically going until we get all of our six images. So it goes 1 2 3 all the way to six images. So if I go here and we'll leave without saving. Uh we have a nanobanana pro image generator. So it's basically receiving all of the information that these agents generated and sending it to uh to nanobanana right and nanobanana pro then generates these images. If I go here into executions and I will go copy to editor so you can see our last run. Uh so

it basically sends this request and I will go show data and sends this request and it generates the image and then it's basically prompted to upload the files into our Google drive and that workflow is then done. Right? So it is fairly straightforward. Again I understand if you've never seen uh N8N4 it might look intimidating. Highly recommend you watch the video. But even if you don't want to learn this You can truly implement this. We really wrote down the implementation for this really step by step. So pretty much anyone can set this up in 20

minutes. You just change the API keys, you know. So what I basically showed you here uh in this workflow, all you will be doing is changing a specific node here for the scrape. So you will be changing URL here. Then you will be changing the uh the URL for the API key for the model for Gemini or cloth. And then of course Your email where you want to send it and then um the specific nanobanana workflow. So again you will need the API key from from Google for nanobanana and that's pretty much it. So we

will copy it set up your N810 account and you are good to go. You know you can start generating images. So I actually wanted to show you a few more examples because when we generate images some of them are fairly easy and some of them are harder. As you can imagine, Something like the honey, it's relatively easy, you know, because it's just a bottle, you know, and Gemini or Nano Banana Pro understands very well what the jar should look like. It's a glass jar, you know, fairly uh fairly common. But some products will be a

bit harder. So, it will might be harder to get it right, especially clothes, jewelry, or art, you know, some very complex designs that you want to replicate onetoone. It might be quite Hard, but it doesn't mean impossible. In some cases, we just want to run the workflow two times, three times, so we get really good images. So, I wanted to show you a few more that are a little bit harder for the AI than the actual honey. So, we ran this for uh dog bandana. So, we did the exact same thing. We copied the asen,

put it into the workflow, and generated it. And I want to show you the result. So, if I go the dog bandana, uh we have it here Again. This is the first image um the first secondary image. We have the emotional hook you know as you can see it's beautiful looks amazing. Then we have the next image. Uh then we have the product quality the material you know built to last. Here we have again the objection reframe because some people might complain you know that it doesn't fit you know they have smaller dog they have

these concerns. So that's why we have this image you know secure not put All day. the cotton holds tight as you can see the the product very actually a lot of detail right even if you zoom in it's quite incredible because this is AI generated right and then then this photo I think this is one of the best photos I generated with AI perfect for every occasion and lastly premium quality everyday price with the patches right and it has the information about the reviews it has the information that it's machine washable because it actually Scraped

the full listing so I think pretty good, right? Uh it did a pretty good job actually. Even though it's cloth, you know, it might be a little bit harder to visualize, it did pretty well even with the with the pattern. The next product I wanted to show you was a coffee machine uh from Amazon basics. Quite a boring product, right? But uh here I wanted to show you the difference between claude the LLM model and Gemini because they actually have a slightly Different approach. The system is exactly the same. We are telling the AI agent

to behave this way. The assignment is exactly the same but you will see that the output is quite different. I actually noticed that Gemini is a little bit more feature oriented, a little bit more robotic, right? This is what I imagine AI optimizing the listings. But cloth is actually very good when it comes to the emotional hooks, right? Because if we if we take a look at the At the honey right here, you know, pure morning ritual. It's it's nothing about the product how big the jar is you know it's just very very oriented for

the for the benefits while Gemini if we take a look again these are all AI generated you know compact design big flavor your perfect morning starts here okay then we have some detail primary use case reusable filter included no paper filters needed like this is great but it's really like feature feature Oriented right so it really depends what are your preferences what your audience responds to because it doesn't necessarily mean that That's bad, right? Because more lifestyle products will be or both the emotions while something very just that's bought for the utility might need to

focus on the feature you know space-saving footprint bruce five cups right so Gemini all about the features fits everywhere and then we have quality you can trust you know the Badges Amazon uh glass caravs but as you can see the product it's amazing right it did a very good job because all it had was this uh sad image and it did a pretty good job from different angles and even the inside, right? Then I double checked and it actually did a pretty good job actually understanding how the product works inside as well. But now let's

take a look at uh Claude because Claude again more emotions, right? Your perfect morning. Simple, Reliable, just right. I I personally like the copyrightiting of Claude way more. Uh can't wait. Don't. Right? This is kind of tackling the objections on the issues that people might have that it takes too long to brew the coffee. Right? So here is the idea that it doesn't take long, you know, nononsense engineering. I also like this presentation of that. Uh sometimes it doesn't do a great job with the fonts that sometimes it just decides to change The fonts for

no reason. But there's nothing easier than just actually taking the photo and putting it into Gemini and you know do the last manual adjustment that you will say like hey change this font because it can do a really good job uh editing the photos. So again, 80% of images will be great or I would run the whole workflow again. You know, I would spend $2 more dollars just to get another variation and then mix and match how it works. Because with these Products, it's a little bit harder since the branding is not as well defined.

If you have something where the branding is just clear like the honey, the listings will be great. But if it's something more generic, we are kind of leaving it up to the AI to come up with the branding on its own. So it might be slightly different, right? So, as you can see, Claude again because her morning routine should be simple, not a science project, right? I I really like These uh right size for real life and for a free brewing with the badges. As you can see, I really really love the design. Again, all

generated what one click. And lastly, we have this oversized sweatshirt because the clothes, especially like these, might be hard for AI, but the way we set this up, I tested it. So, I'm going to show you how it handled this hoodie. Um, ultimate cozy warmth like wearing a warm hug. So, if you can guess if this is cloth or Gemini, right? Of course, this is cloth. Uh, ultimate cozy warmth like leaving a warm hug and move freely. Stay cozy. Again, very nice. And as you can see, the product is actually shown very well. It's the

same color, right? It's beautiful. And here, double layer luxury. Like even these details, the stitching like just look at that. It's it's really uh it's really amazing. Then we have oversized by design. We are tackling the objections from the Reviews. Some people complaining that it's too big, but we are saying like it is supposed to be big, right? It's not a defect. It's not a problem. We have one sizefits-all. And here it actually added a little puppy, which I approve. And perfect for indoors and outdoors. We have the same issue with uh the font I

believe. Uh no, actually these are fine. Uh perfect for indoors, outdoors. and then it chose to do it as a perfect gift because likely a lot of people here Mentioned in the reviews that they buying it as a as a gift for someone. So this was a showcase the same workflow just using different asens that it was able to handle very different niches right jar of honey consumable product uh then we had this bandana we had coffee machine and we had this oversized hoodie and as I said if you want to use this workflow you

can make a copy I will leave the link in the description and just to make it easier for you we also Included a guide how to set this up it's free you know uh that you can just open this nano banana pro and then hit automation you Google, copy it. Uh, and then there is literally step by step how to set up epify, how to set up your Google AI studio, how to set up N8N, everything here is in detail, right? So even if you don't have any experience with N8N, you will be able to

replicate this in your business and start using it. But of course, what offers here when It comes to uh when it comes to creating these uh automations and NAN, it's not just about this one. You can change the system prompt and start generating images for your social media, right? You can start generating content for Tik Tok, right? Because once you unlock the power of AI automation and you understand how all of these pieces work and how how you can combine these tools, it is incredibly powerful. This is just one little thing that I just showed

you, But you should really understand these tools because it's only going to get more powerful uh going forward. Now, we are going to replace your spreadsheets with custom AI apps. Most sellers are still trapped in massive Excel files that break constantly and take hours to update. But we are going to build clean, simple tools that do exactly what you need. We'll use Lovable to build these apps, which means that there is no coding required. You Describe what you want and AI builds it. We'll create real apps from scratch, a search term report analyzer, a review

insights dashboard, and a listing generator. So by the end of this section, you'll know how to build custom AI tools for any workflow in your business. So let's dive in. This is a quick overview of the AI apps we are going to build today. One of them is review insights. This app allows you to analyze hundreds and hundreds of reviews With AI and give you insights about your products, the strengths and weaknesses based on the feedback you received. So as you can see, the strengths and weaknesses all broken down nicely with the quotes from your

reviews. It will also give you ideas on product improvements, how to prevent the negative feedback uh you received on your reviews. It also gives you listing images again based on the feedback you received solving the pain points. The Next one is Amazon search term report insights. This is an app where you can upload your Amazon search term reports to give you a detailed summary about the performance, where are you wasting your ad spend, what are the winners you have in your report. So again, very useful. And the last one is Amazon listing copy generator. So

we upload a listing uh keyword list into our app and it will generate ideas for images uh A+ content, but it also generates the title for the Product. It generates the bullet points for the product and the backend search terms. So this app will generate your keywords. So now we can go ahead and build these three. This is a live build of our AI apps. We are going to build all of this from scratch. So I do believe that microAI apps will soon become an essential part of our businesses and they will replace most of

our spreadsheets for reporting and other functionality. Because if you are like Most sellers, I assume that your Amazon business is currently powered by dozens of complex spreadsheets. And it makes sense because we can either rely on third party software, which in many cases doesn't do exactly what we need, or we can build spreadsheets. And as you know, spreadsheets, especially the complex one, have their own issues, right? They break. It's hard to teach someone how to use them. It's hard to update them. So in this case, I do Believe that micro apps we can build with

AI actually solve this problem because they are essentially mini tools solving a specific thing in our business. We can develop in just a few minutes without learning how to code or having a developer. So you can either create dashboards or you can create apps that fit your exact workflow, not generic software. Right? So we kind of flipped the script because before with third party software we kind of had to Develop our workflows around the software but now we are actually building a software that reflects our workflow and our framework. It basically reflects the way we

want to do things and I do believe that this is significantly better than spreadsheets because the interface is cleaner so it's easier to train people. There are no formulas you need to learn so it's less likely to break. Plus, most importantly, we can actually integrate AI into it if It's supposed to serve a specific function. So, we can basically turn a dashboard into an app. And even though it may sound intimidating, all of this is actually easier to learn than spreadsheets. So, how you can use this in your business? We are going to show you

three different apps. One is search term report analysis, a very simple dashboard. Then, we will create a listing generator. essentially an app that will generate full listings out of A list of keywords. And lastly, the most impressive one is actually review insights where we will scrape reviews from a listing and it will give us insight about the products, about the listings, how to improve that, how to optimize it, right? So, this one will be powered by AI. I want to stress out that this is not like a cool tech demo. I really want to show

you practical use cases, something you can start using in your business right away because out There there are many like tech demos that are very very impressive, but they serve one specific purpose. It looks nice on the video, but it's very hard to make it work in the real environment in your business. So, I really want to position this as an operations upgrade, something you can really start using right away. So when it comes to building these apps, I will cover my process, right? We are going to work with Lovable, which is essentially an uh

a Software that helps us build the apps with AI. So we will describe what our app should do and hopefully the AI will then build the app. Uh but to make this even easier, I use CLA, right? a large language model to write very very detailed descriptions of the app because lovable works in a way that we need to describe the app of course using words and it's not always easy to cover everything you need in the first prompt that's why we use something like cloth In my project right here which I created uh lovable

AI dashboard bu dashboard builder because that will help us write the instructions. So as you can see you know it will write the detailed app specifications it will write the API integration guide everything will be created by cloth. So first of all we have a conversation with cloth we clarify what the app should do what it should look like all of this right we will have a bunch of documents and we Will then put it into lovable and hopefully create the app on the first prompt uh in cloth I have this project with a detailed

system prompt for actually building these dashboards. So that will be available in the link below. So you can access this prompt you know get a copy of this uh to help you create these dashboards because if you don't have that uh sometimes cloud doesn't really give you everything you need. So this one is already trained and Our goal here uh is to essentially oneshot the app. It means that we want to create the app or at least 90% of the app on the first try. Right? that will save us time and also headaches because every

iteration of the app can uh potentially make it more complicated. So we really want to make sure that the description is as detailed as possible on the first try. Once we pass it into lovable then we will need to iterate. There will probably be some minor errors But I will show you how easy it is actually to fix those errors with lovable and we will be done um in 10 20 30 minutes building these apps. So as you will see it is actually way faster than building spreadsheets and it will do exactly what we need.

I also want to stress out that we will be building simple apps app that solve a specific problem right it does one thing extremely well. I see a lot of people try to build something very Comprehensive some all-in-one tool like you could potentially do that you know your helium 10 your seller port all of these things but we don't want to reinvent the wheel right if you have a profit dashboard that works like sellerboard or whatever use that one don't create new one but if you need some specific function you know a framework following your

framework to create listings that's where this shines like very specific purpose don't think Multi-purpose allin-one one simple app one specific specific purpose. So you can have multiple micro apps, right? That's why we call it micro apps. So it will be faster to build, easier to use and zero maintenance. Beautiful. Now, so the workflow philosophy behind this, how we are going to structure these apps is that they will work like this. You will have some report, some CSV, either you get it from Amazon, keyword research, whatever. And the idea is that we want An app where

we will up upload the report and we will get some result, right? the app will visualize it. uh it will use AI you know to answer questions to describe it whatever we need right so in this case we upload keyword list and we get full listing on the other side the title bullet points backend search terms and all of that stuff right so this is the idea this is the workflow because I want to stress out that there are essentially two ways we can do it we Can do like proper apps with databases right where

you would care about the login you know this is for other end users you know if you want to develop essentially a software. So you would have some automations that automatically download the reports in N8N. You would have a backend database that would store, you know, you would have history of your all of the data sets, right? And that you would have the front end in lovable. We're not going to do this, Right? This is advanced. So you can essentially create an app that automatically pulls the data. It automatically stores them and visualizes them, right?

So essentially this is how you would recreate something like a sellerboard. not going to do that. What we are going to do, as I said, our workflow philosophy is simple. We just want to show and work with specific reports. So, we will have a CSV report file, you know, your Excel file. We Upload it into our app and we get some output. Very straightforward, but it's 10 times easier to build and it does like 80% of the work we need, right? So, let's get into it. We can start building this. And as I said, we

are going to begin with a search term analysis. So in this case, this is a simple dashboard. We are not we don't need AI implemented into this dashboard because essentially we are using formulas. Um we are using formulas to calculate everything we Need. So here I always prepared the assignment prompt just kind of describing what we need, right? So I don't want to waste your time. This is what we will uh paste into cloth. So the idea here is that we will have a CSV report we downloaded manually from Amazon, right? So we have some

complex CSV report and we will paste it into our app we create in Lovable. That should hopefully give us all of the calculations, recommendations, the Filters, insights, all of that according to our assignment. So the assignment is simple. We will attach uh a sample Amazon sponsored product search term report and I want to generate a lovable app. Right? The idea is that uh the lovable app will receive the CSV file, right? Essentially what I just described and the output will be divided into cortabs. Executive summary, detailed performance breakdown, wasted ad spend, winners overview, uh and

data explorer, Right? Before analysis, users will configure performance targets uh target a cost, target rows, minimum spend dashboard, minimum clicks, keep the recommendations brief, and so on. So, as you can see, this is a simple assignment, right? This is kind of what I can write by myself. But this wouldn't be sufficient for lovable because as you will see if I paste this into our cloth project, it will ask us a lot of questions and those questions are Essential. So we can go deeper. We can add as much detail as possible and you will see what

cloth actually does with this. So we will just take this assignment right that is the first assignment you would write. I would go into the lovable AI dashboard. As I said, make sure you create a project in your cloth. In the instructions, you copy the system prompt which will be available in the in the AI agent vault uh uh down below. So now I copy pasted The this prompt uh into Claude and I also included the search term report as a sample here so he knows exactly what those reports look like. So now we will

send it to cloth and you will see that we will get a bunch of questions and as I said I will probably make a cut in the video so um I can answer all of them. Now he asks us multiple questions you know about data and calculations. Then we have recommendations and actions UI details technical and flow and so on. So We have 16 questions. So I now will sit down and actually answer all of these. But as you can see, even though the assignment I thought was quite, you know, detailed. I included everything I

wanted. There are a lot of unknowns. And this is why we do this first in Claude. So we can be as specific as possible. So let me now answer all of these questions. I answered all of these questions right here. And now he's asking for confirmation summary, right? So he has everything he needs and he's just asking for confirmation. So this is how we trained uh the particular project. So I can say yes, proceed. And now he will generate all of the files we need and the prompt we will then copy paste into lovable. And

now we have these beautiful instructions. So he says copy and paste this into Lovable. And he also generated the files. So as you can see we have the detailed app specification file and also Amazon PPC Best practices file. So I will now download these and then we will upload all of this into Lovable according to these instructions. Now we are in Lovable. Of course, in order to build this, you will need the premium account, but it's like $20 to buy the credits. And with the credits you get for one month, we will be able to

build all of these apps. You can then use as long as you want. So, it is definitely worthwhile to invest into that. So, I Will copy and paste the uh the prompt right here and then I will click attach where I will include all of these files. So, we will have the detailed app specification, the best practices, and also our search term report. he will need. Right? So, we will wait a second to upload it. The prompt is here as intended. And now we will see what Lovable actually does with it. We waited for roughly

5 minutes and Lovable actually created the app according to Our assignment. So, as you can see what I created, design system, core features, everything is we requested here. Right? That was our first prompt. He did this for three, four, five minutes. And now we are ready uh to try this. I want to stress out that it rarely ever works on the first try, right? We will likely have to make some adjustments and I will also show you how to do that because it's very straightforward, right? We are just talking to the AI the same way

you Would talk to chat GPT or clo you can just talk right here and he implements the changes. So let me see Amazon PBC report analyze upload your Amazon sponsored product search term. So I will choose file here. We have the search term report configure performance targets right as as we requested and as I said before you can make this exactly you know reflect your workflow so this is just me creating it you know as I'm thinking how I would create this app but Maybe you have different goals maybe you want to structure it in

a different way so you can do this easily right that is the power of it you can customize it so analyze report let's see what comes out and now we are noticing we have the ad we the app but we're noticing that total sales here is not showing right I know that in this report uh there should be some sales so we have total spent works but total sales are not showing analyze search term 7,200 Uh performance distribution recommendations all otherwise it looks pretty good right but as we can see sales here are zero everywhere

so a cost and everything is not showing correctly Right? Everything else should here work as intended. So we I will just go here. But if I would write something what I want him to do, he will do that immediately and it would start eating my credits. So what I can do, I can ask that I want to just chat, you know, chat Without making edits to your project. So I would click on that and say I noticed that the sales section is empty showing zero. So I simply ask him that I notice that this section

is showing zero and he should look into that right using the chat feature and as you can see he then goes takes a look at the CSV file and notices that there is some issue which to be honest I don't understand at all but I do not care as long as it works. So then he asked you to implement The plan. So I said like go ahead implement this and we will see if it works or not. So now we implemented the changes and we can upload the report and see what it looks like right.

We adjusted the app to showcase everything. So now we have the target a cost all of these thresholds are here. We will click analyze report and now we get this overview where we see the the total span, total revenue, overall rorowass, we have actions, recommendations, right? So everything we specified in claude is now right here. Um detailed performance. So we can just filter this out so you can see all of this, right? So, it's showing uh the key metrics for all of our 8 and a half thousand search terms. We have wasted ad spend. Of

course, the data is anonymized here, but you can see, you know, critical high spend, total wasted spend, you know, 8.5% of total budget is wasted on uh search terms that are not performing, you know, Not getting any sales. Um, then we have winners, top five rows performers, top five revenue drivers. So you can see based on our specification we get something like this and the recommendations. So I would probably spend some time you know adjusting this according to my framework what I'm actually looking for but you get the idea right it is fairly straightforward building

an app you can spend 30 minutes 20 30 minutes and you would build Something like this and then every time you download your report you can put it in there and you see everything you defined that you want to see. Then you can simply publish the app. I will again give you access to this. So uh in the in the link in the description so you can make copy of the app, you know, make your own changes, start using it and kind of get the idea how all of this works. Plus you can build this

only for the internal purpose, right? If you Publish it and then you have a link, your team can start using it, right? It's your own app you can use internally. So now since we have the Amazon search term analyzer which was this one, we can continue on the next app and in this case it's going to be an AI app because we are going to use AI inside the the app we are going to create. So the idea here is that um we will have a keyword list you know so you will do your manual

keyword research However you do that we will have a list of keywords for our let's say new product launch and the idea is that we will upload the keyword list into lovable lovable will send an API call to clot right so on the back end it will reach out to the AI to help it transform the keyword list into an optimized listing and then the final listing will be presented in Lovable. Right. So, first of all, in order to get the uh API call, you will need to set up your cloud Console. So, you will

simply go cloud console and you will log in in there. You can do this with CHGPT, right? You can get the API key from CHGPT as well. So you would go you know chatgpt console and you then need to generate your own API key because this is different from the cloud we are using here right this is used for the backend software and here the beautiful thing about this that you are charged per usage right just to give you an idea about the costs Involved so building this app would cost me like two to three

dollars right roughly on the credits we used on the search term report analysis and here for generating the uh the listings you will see that we will pay once essentially to lovable like a few dollars and then each listing we generate is just a few cents right so it's basically basically for free but let's begin right uh again in our database here in listing generator we have the keyword list right as I said So we kind of already did this manual step and now we will go back to claude and we will copy paste this

assignment to claude so this assignment again kind of just breaks breaks down what we want from this app. So we will upload the list of keywords and we essentially have the output divided into core tabs strategy breakdown highle summary of the conversion rate optimization strategy for this product positioning and so on. Again if this was my brand I would Hardcode you know our um our positioning our target avatar and all of this into the app. But since I'm not doing this, I will just say like, hey, let the AI decide how to optimize this product

and how to position it, how to sell it. Then we will get image ideas, you know, detailed description of all secondary image with reasoning. A+ content include ideas of what to include in this section. Title include three variations. Bullet points include three variations. Backend search terms, list of search term uh in to include. As you can see again very simple assignment but what we are going to do is we are going to copy it and paste it again into uh into Claude and we are also going to include the file as a sample so he

knows what the keyword list looks like and again Claude will ask us a bunch of questions we need to answer in order to get the detailed assignment for lovable. Same scenario as before. Claude is asking us A bunch of questions so he can specify the uh the assignment as much as possible. So again I will take a few minutes to answer all of these. Right now we answered all of these questions. We got the summary and Claude is asking us for confirmation. So now we will say yes proceed. So he can generate the prompt. We

will copy paste to lovable and all of the the specification files. The full assignment is ready. So the same as before, we have the detailed app Specification file. So let me download that and put it into our listing generator, right? So the name will stay the same. We also have the cloth prompt template. This is the one that Lovable will be sending via the API call to AI as I said. And here, copy paste this into Lovable and attach the files listed below. Right? So the instructions are very detailed. So what I'm going to do

is I'm simply going to copy it. Go back to lovable copy paste the the prompt Right uh listen copy generator so as you can see all of this attached files now we'll go attach uh cloth prompt template detailed ad specification and martini keyword list um so open now we have the three files uploaded successfully so we will go ahead and send it to uh to lovable and in the meantime as lovable is working on this we will go to the cloud console And as I said, uh you will have to upload like $5 into your

wallet. So the API keys can then just, you know, Consume some of the money you have there. So you will go to settings and billing, you know, and top up your card. So as you can see here, I topped up like $5. We still have $3 left, which is more than enough. So we'll top up like $5. Go to API keys, and we will generate an API key when Lovable asks for it. And we will just copy paste the key into Lavable so it can send the call uh to claude and get the full listing

as an output. Now Lovable is already asking us To enable the Lovable cloud. This means that he will allow us to input the API key and store it securely on their end in our database. So we will go allow. And now it will likely ask us to include the API key when requested. So let's give it a second. He's asking us to add secret and as you can see entropic API key. So I'm going to go here create key and I will go you know lovable listing generator and I will click add. It will generate

the API key. We will of course Blur it. I will copy the key. Close this and I can then see how much the specific app actually spent. So we will be tracking the cost right here. I will go back into Lovable. Paste my key. Go submit. Now he has everything he needs. So he will start working on the app. Perfect. So our first version of the app is ready. So as you can see complete features, CSV upload cloud haiku integration six result tab. So everything we already described in our Initial prompt is created. Uh but

as I mentioned rarely ever it works on the first try. So let's upload our keyword list right here and ask to generate listing copy and see what happens. So, of course, as I said, we got an error, but I don't even have to investigate. I would say try to fix, you know, immediately and it kind of uploads the error back to the AI so he can take a closer look at it and help us fix that. So, what I really like is that Lovable Is able to identify the issues, right? So, we said, "Fix these

issues and he said, I can see the issue clearly. The model name I use doesn't exist." So, he fixed that. He used claw 3.5 haiku, but then I told him like, "Hey, I actually want to use the 4.5." So then he implemented the 4.5. So let's try it again. Uh it is likely that we will get some error again. But the iterations are part of the process, right? So we will like have to do that three four times. It will take 10 minutes to actually go through this before we get this to work. So let's

see what happens now. So we got an error as expected. So we will try again fix these issues and Lavable will look into that. hopefully fix it and we will s soon we'll be able to actually make this work and really you don't need to understand what these issues are because you always click on fix these issues you know he finds the issue uh fix it and here you can see claude is Returning the JSON wrapped in markdown code like you don't really need to know what's happening but we will try it again and hopefully

this time it will already be all good beautiful it worked so now we have generated listing copy and as we requested we have these sections strategy actually let me zoom in a bit Um, we have strategy, images, A+ content, titles, bullets, keywords, right? So, the strategy breakdown. Uh, if I did this for our business, I would Of course include and hardcode these things about our customer avatar, our positioning, all of this into the app. But this time, I let the AI to determine this, you know, the the whole strategy based on the keyword list, which

of course is not optimal. But he breaks down the strategy right here. what he does, you know, the keyword hierarchy, the target customer profile, you know, private cu primary customers, adults aged 30 to 55. Uh, it's Martini glasses, Right? Who entertain regularly or aspire to sophisticated home entertaining, they value both aesthetic and functionality, conversion rate optimization strategy should emphasize setbased purchasing, you know, and so on. So, he kind of breaks down the strategy he's going to use for the listing, keyword integration, competitive differentiation, keep psychology triggers to leverage, social proof. So he actually breaks down

the whole Strategy for the listing and then we can go into images, right? So image number one, complete set of four in context. So that's high priority, right? Show all four glasses arranged on an elegant entertaining surface. Marble County top parkard as you can see a lot of detail and also the reasoning behind it like how it's actually connecting to the strategy we determined, right? Addresses the high highest intent keywords. Uh, this hero image conveys the complete Entertaining solution reduces purchase anxiety by showing the full offering. Then we have size comparison and variety options, close-up

crystal quality and craftsmanship. You know, always the reasoning, right? Implies quality positioning tied to brand searches and so on. Entertaining scene with multiple cocktails, description, reasoning. So, it's very very good even though we didn't spend that much time actually specifying what the AI should do. Um, Give presentation packaging and so on. So you have the ideas for your secondary images but we also got the A+ content you know comparison chart feature breakdown all of this is here. So based on this we could create the brief for our photograph or for our designer we have the

titles um premium crystal martiny glass set of four elegant cocktail drinkware for entertaining again with the reasoning feature Specific. So he we asked to create three different variations of the title you know with different kind of angle. So this one is set focus, benefit driven, premium positioning. This one is feature specific size and variety emphasis brand and versatility focused. The same thing he did for bullets, right? So he created the bullet points according to this strategy with reasoning. And lastly we have uh the backend search terms. So now I could Upload these uh keyword lists

for different products and he would determine this. If you have, you know, one, two brands, you would hardcode all of these details into it. So maybe it would be a little bit more specific, but you get the idea. Now you have the app. I could click publish. My team could use it every single time to quickly generate listings, right? And we could then potentially automate this even further, right? So we could connect it into our Database where we have the keyword list so we don't have to upload it manually and so on. So this is

again very useful to have everything in one place all generated with AI. And now the coolest part is that this whole thing cost us uh 9. You can see 0.09 right? So pretty much nothing. So that is the beautiful part. Now we have the app and we can use it again and again. Um, now we can proceed to create the last app and I'm actually the most Excited about this one because it's extremely useful. We will be creating an app to get review insight. So, how this is going to work, we will get uh a

scrape of Amazon listing the the reviews. Actually, I'm not going to cover this in this video because it's outside of the scope of this video how to actually do these scrapes. But you can use Bright Data or Empify uh to scrape the Amazon listing to get the text of reviews on a specific listing. What you're going to get is a sheet with all the reviews, right? So, let's say you have uh 200 500 reviews in one spreadsheet. We will copy paste to Lovable. Lovable will send it to Clot exactly the same way it did for

the listing. Claude will analyze it and give us final insights for um from the reviews. So as you can see we are asking Claude to create five core tabs you know summary key insights from the rest of the dashboard. Then product strengths And weaknesses what people like and dislike about the product. Product improvements ideas how to improve the product based on the feedback for listing section ideas for images to optimize conversions based on the feedback. And lastly, data explorer so we can go through the data. So let me put this into cloth once again into

our lovable AI dashboard. I will upload the file uh review insights the full review sheet and again claude will ask us Multiple questions to finalize this dashboard. If you are not sure about the scraping, I actually have a video about setting up MCPS. I will also link it down in the description which allow you to use cloth to access this directly like epify to scrape the listings. It's not complicated right data or epify you will be able to get your um review list like this uh very straightforward to do that and then you can create

these dashboards. So let's see we will get a Bunch of questions again I will answer all of them and we will continue. We answered all of the questions and now we have the confirmation summary, you know, breaking down everything we asked for. So we just, you know, specified everything in detail. So now we say yes, proceed. So he creates the prompt and the files we need, we will then copy and paste into Lovable. Now we have everything we need. We have our opening prompt for Lovable. So let me copy that. We have detailed app specification

and cloth prompt template. So let me download it. Put it into our review insights uh sheet uh folder. So here we will put this also the cloth prompt template. Go back to Lovable. Copy and paste this prompt. And now we're going to attach all of these files from the review insight. So open. Now we have the full assignment and we can send it over. So Lovable can start creating this. And in the meantime, we Can create another API key for this app. So this would be lovable review insights. And we will create this. Copy the

key and wait for lovable. Now we will enable the cloud the same way as before. So we can input our API key so it can securely store it and continue building this app. The process is exactly the same as before. So I will include my API key right here. And now Lovable will build the whole review inside dashboard. Few Minutes later, our first draft of the app is ready. But as we know, it never works on the first try. So let's upload it and see. So we have our review sheet here. Then we can take

analyze reviews. And we already have uh an issue right here. So we will ask Lovable to help us fix it. We ask Lovable to fix the issue and implement the plan. So hopefully this time we will get a bit closer again. Uploading the file another test run. Actually it was the same issue as Before that lovable decided to use the 3.5 haiku model from claude which is the old one. Uh but we want to use the 4.5 because there is basically a token limit right we are sending a lot of reviews for analysis. The older

model couldn't handle it but the new one can right? But again labable helped me uncover this issue. It's the same issue that happened before. So now it's implemented. We can select the file and try the analysis once again. Our analysis is ready. So This time it worked well. So as you can see we have the summary you know the sentiment overview uh the top themes you know what's actually repeating in the reviews exactly as we asked. Extreme fragility and breaking theme feeling glass beautiful modern design. So we can see what's positive what's negative the recurring

themes and how often they actually appear in reviews. There are some quick stats. 68% of reviews mentioned glass breaking easily or Stemming from fragility uh priority actions. Right? So this is how we structured the the summary. But the most impressive part is actually the strengths and weaknesses. Right? So we see the strengths uh where our product shines. You know modern, elegant, unique design. We even get some quotes from the reviews kind of supporting this claim. Uh and we see how many how many of these reviews actually mention this. 89% of five star reviews highlight design

as Primary reason for purchase. Design mentioned in 62% of four-star reviews, you know, excellent value and aesthetic appeal, perfect for gift giving and so on. So this is a gold mine, right? This information is a gold mine to actually see what customers like, what we should highlight in our listings and also what they don't like, the weaknesses, extreme fragility and frequent breaking. uh paper thin glass. You know, 94% of onestar reviews, 87% of twostar reviews Uh mention breakage. Even 11% of five star review mention breakage. Then we have paper thin glass feels like plastic, widespread

manufacturing defects, high damage rate on arrival, dangerous sharp edges when broken, unrealistic product lines versus delivery. And as you can see, so these one are the most recurring one. We should pay to these attention. But now product improvements, right? Based on the reviews, we can see what the AI Recommends us we can do to improve the product. You know, increase glass thickness and upgrade material quality, redesign and reinforce stem structure, implement rigorous quality control, upgrade shipping and packaging to absorb impact. You know, so it's what to change? It's very detailed, right? Improve protective packaging for

a fragile product. Why this matters? 35% of orders arrive with damage despite existing styrofoam packaging. You know, Supporting evidence. you get the quotes from the reviews. This is a gold mine for every seller for every product. So, as you can see, it's very very detailed all of these um all of these ideas how we could improve the product. But that's not all. We also have listing images, you know, ideas how to consider these pain points, the the strengths and the product improvements, like how to put it all together to create an optimized listing, you know.

So we have main Images uh lifestyle you know four images paper customer need to see how glasses look in actual use elegant but safe handling um visual focus callout text perfect for celebrating life's moments uh customers want to see versatility can these be used for different occasions you know multiple style scenarios champagne tosses wedding procco branch motor home bar display party setting it's amazing right infographic uh different types of images comparison Customer want to verify this is actually glass not plastic. So now it's addressing the the issues that people might read uh might read in

the in the negative feedback right very very valuable. So all of this on one analysis and again as I showed you before how much did this analysis cost so let's see $0.06 $06 for one product. Amazing. And now we have this app. We can use it for every single product. Right? Beautiful. Of course, um there are some Improvements I would make to this app. I would automate the listing scrape. Right? In the beginning, I would just add a different section so we don't have to manually download the CSV. We would just input the ASIN, click

analyze. It would scrape the listing, pull the pull the reviews, send it to AI to make it a little bit more convenient to use. So if you want to see this, I will create another video to actually make this improvement and share this app with you. But really, I was blown away the first time I tried this, how good this actually is and how well presented it is, right? Because now everyone can use this, right? Just click of a button and you are good to go. You can publish it.