The YouTube Parasite System (ChatGPT Growth Hack)

What's going on everybody Nick here got a great system for you today where we're scraping YouTube videos we are then going to be processing their transcripts then we're going to use AI to come up with new and interesting twists on that content um before ultimately repurposing that into a bunch of different content formats including Facebook Instagram LinkedIn uh Twitter we're even going to create a new YouTube script based off All this let's do this thing all right so we're going to build this in a sec but I want you guys to know that I haven't

actually done this specific system before I'm going to be building it alongside you live so you're going to see the detours that I take along the way you're going to see my thought process is narrated as an automation developer and the reason why I do this is because I want people to see what an actual development process looks like I Don't just want them to see like the pretty final picture so to speak I think it's a lot more informative this way so yeah um I have an idea of how I think the system is

going to work but I don't entirely know whether or not the system is going to work and we're going to figure it out as we go along um Tobias here said would something like this be possible for YouTube meaning scrape transcripts of new released videos of channels we want to use as a starting Point and then have it create post off the material so when he said that uh my first thought was probably and uh I spent about 5 minutes just thinking through how I'd do this um and and here's here's what I've come up

with so this is just my tentative outline I don't actually know 100% if this is going to make sense but we're going to we're going to figure it out as things go along usually what I'll do is I'll like start with some end goal in mind And then I'll just work through three or four possible variations of how I might do it um this process takes me 5 minutes and then you know I end up with some list here so what I'm thinking we're going to start with is we're going to scrape some some YouTube

videos obviously because that's going to be the data source we're then going to get the transcript of the YouTube videos I don't actually know how uh I'm going to go from from here to here to be honest Because uh some scraping sources and some resources that you use they actually include the transcripts built- in um other others have some sort of like buggy functionality here and then others I don't have that at all so depending on which one we use we may get the transcript if not we may actually have to like transcribe the text

using some AI platform um what happens after is I'm thinking we're going to summarize the transcript then we're going to check The summary to see if the content fits our filters so like you know if we're creating content for curating content really to be repurposed on our own Brands it needs to fit our brand tone of voice it needs to fit sort of the things that we talk about so what I'm going to do is I'm going to set up an AI prompt that looks at some examples of maybe past content that we've done or

uh I don't know just like funnels the content and ensures that any new content that Comes into this funnel or queue or database um is going to be something that we would actually post so that's it's going to be pretty cool after we're done with that uh we are going to store it in a database with a unique ID the database in this case just going to be a Google sheet so very straightforward anytime that you hear database you can just basically replace it with like Google sheet and 99% of the time you're going to

be fine so that's What we're going to do next and then if new we're going to use AI to turn the transcripts into highquality structured posts for various platforms I don't entirely know what platforms just yet but I'm thinking we're probably going to do uh like I just want like two fundamentally different kinds of platforms here so you guys could see I'm thinking we're going to do some sort of like Twitter pre-draft I guess it's called X Now um or uh we'll we'll also do something longer form so maybe like uh LinkedIn and then I

think on top of that and really this is just like to demonstrate the the value of this we're also going to um generate uh like an like an outline for our own video and I'm going to try experimenting with um maybe like a script or something I'm going to try experimenting with an I don't know if this going to work but I'm going to try Having this model that I pass it through come up with something unique that it can add to this because I don't just want it to be like a one for one

sort of plagiarism thing I've gotten a couple comments as to a few of my previous posts where people are like well like you know is this just copying the guy's posts and I think that's a that's a reasonable claim to make now don't get me wrong I don't really care much for the morality or ethics behind Repurposing publicly available Twitter posts people are going to or sorry um YouTube posts people are going to copy anyway you should see the number of people that have copied my videos I'm not worried about that um this isn't like

a you know like like a moral thing I just wanted to show you guys how to actually like make the content even better maybe add some new twists to it add some some value um and then after all said and done we're going to post These on our personal company pages so this is what I think is it's going to look like um this is not actually that complicated like of a flow to be honest we'll probably get this thing done in like half an hour to an hour unless I am fundamentally misunderstanding something but

yeah I'm just going to show you everything that I'm going to do along the way so the very first step about scraping YouTube videos there are a variety of ways that you can scrape YouTube videos the way that I'm probably going to use and if you watched my last last three or four live builds you're probably like oh boy what's he going to say next um is it is it appify and yeah it's going to be appify and I'm not saying this because I want you guys to only use appify just because it tends to

have the vast majority of scrapers um that I use for like this sort of content repurposing purpose so appify is a Content uh sorry it's a scraping Marketplace which has hundreds of scrapers developed by mostly real actual other people Like Ur that do things like extract data from Tik Tok uh get data from Instagram you know in our case we're going to be doing YouTube so scrape videos from YouTube scrape shorts scrape comments I've done the comment scraper um as part of my Approach that's actually what I'm doing right now like I'm I'm doing this

video because I used a comment scraper to scrape my comments And then pass that through Ai and tell me if this is something people wanted there are hundreds of not thousands of different scrapers here that you can use for basically any purpose and because it's a Marketplace people tend to compete on cost and stuff which keeps it reasonably cheap for you so uh I'm realizing this says LinkedIn parasite system let's just go YouTube system must have duplicated that from a template uh so what I'm going to do first is I'm Just going to look for

scrapers that are built uh like pre-built for me rather than woring about making my own if I had to make my own it might take way longer it probably be dozens of steps here in terms of programming and rate limiting and I need to you know proxies and and like that appy just takes care of all that stuff for you so my go-to anytime I have a new scraping project just start with something that's that's pre-built if it does 90% of the job for You that's probably sufficient so I'm just going to open up like

three or four of these and the reason why why is because I just want to see um I guess we want the video scra right I want to see we got a caption script right that's perfect if if you know I can just get something low cost I want to like compare these with each other okay this one gets the transcript that's perfect if we can get the transcript from videos we are freaking Balling here oh that one looks cool topaz Shing gun my Naruto Bros uh okay cool let's double check this so this is

a YouTube script you pay $5 for a th000 videos so what do you get you get videos titles met description metadata that might include the transcript I don't know I'm just going to command F and type transcript I'm not seeing anything here so I'm just going to go to my next tab then I'm going to type in transcript again I'm not seeing anything This a short scrapers that makes sense I clicked the wrong one this is subtitles caption scraper transcript yeah I see one over here too I'm going to give a click um this looks

like a subtitle right here so this is this looks like workable author app IFI video description so are these different videos if these are different videos then Yeah are these all the these are all the same oh you know what it scrapes one video but then it has all of the subtitles here okay well I mean this is workable like we we might be able to work it and it's like 15 bucks a month so if this works we are golden um okay video URL or ID so I guess actually if you think about it

we we kind of have to go back one step we have to get a list of videos that we want to scrape right so not actually not actually just The transcript so my system's going to have to be a little bit more complicated um let's just do like a let's do a step zero here uh where we scrape list of videos to scrape and then I'm just going to drag that stick that on top here and and did kind of with my whole perfect flow there cool uh so yeah we are actually going to have

to get like a list of videos now I think about it dumb but anyway uh from here looks like we Have three or four scrapers we could use I'm I'm probably just I'm just going to jump into like whatever one and then see how things go um all of these cost a little bit of money I I think I'm probably just going to start with uh sorry there was a transcript here right no so I guess I'm going to start with h we just have to make a call so This one's a try it free

no credit card required I don't know if it's 100% actually free but I'm just going to give that a try and the very first thing I want to do is I just want to verify whether or not this actually can generate a you know uh video transcript so hold on one sec cuz my video is going to play when I go on my Channel all right sweet I stopped it two mes would be too much and for the Purpose of this why don't I just get like a shorter video I don't want to get yeah

let's do this one back before I understood how headspace worked so I'm going to I'm going to share this I'm going to get the share link I'm just going to open this up in a new tab but I'm going to um turn all of my audio down hopefully this doesn't play and the reason why I did that is you know you saw how the link changed there at the end right so YouTube uses Um different links for sharing versus uh which tab am I in this one here uh versus um actually playing okay so now

that this is ready I'm just going to give this a go and we're just going to see if we can scrape the transcript generally I always like to start with like the core data that I'm going to be manipulating in this case it's obviously going to be the transcript so we just wanted to double check and see whether or not the transcript can actually be Scraped with this actor um if so then we'll continue using it if not then we won't so I'm seeing a bunch of cerilla so pollution LEL so I don't know I

imagine this is Russian my Russian ain't so good I think uh this is basically just telling us that we can't because there's some problems so this actor is probably not going to work let's move on to another one here video subtitles caption scraper I'm going to try this one for Free it says I'm going to delete oh it's cool that we can feed in a list of videos here it's going to be helpful uh language optional can we just do auto yeah know I think we're just going to do we're going to fetch the autogenerated

ones cuz that's totally fine for us and then I'm going to paste this in and we're going to click uh start it's going to tell me that I have to rent this we get two free days so it's not going to cost me a buck to do this so I'm just Going to go okay and then save and start let's see what sort of data we get from this actor um and that will inform us as to how we're going to proceed with the rest of this now because it's one you know video This should

run reasonably quickly you can see I'm also being build for usage so I've been build. 2 cents. 3 cents so far um a good way to see if the actor is actually working is just going to the log and just seeing if there's tons of Issues and errors here so I'm not seeing any issues I'm going to go back to Output I don't know what happened there okay yeah great okay now we have good God this thing's huge can we just get like one thing with the whole freaking no we can't that's sort of silly

um I mean ideally we just get one thing with the whole transcript this is giving us like the line by line it's not that big of a deal we could Just um join the whole thing into a script inake which is the platform we're obviously going to be using so yeah it's not that big of a deal I think I'm just going to run with this for now um now I'm going to put this into make so I I just want to verify okay we have the data in appify can we get this data to

my no code platform because my no code platform is going to I'm just going to call this like YouTube scraping system um because you know it's one thing to Have it in appify it's another thing to like actually be able to automate it so to do this I'm going to go to get data set items under the appify module I'll go back to my appify link and then what I want is I want the storage Tab and I want to copy the store ID here sorry um go to data set make sure this tab is

a data set copy this data set ID paste it in here I'm going to sign into my um my account because I have multiple connections and then I'm just going to Return all of the things so 100,000 just to be safe I'll click run this module only we're now connecting with appify and we're getting this so as you can see this is quite a big like quite a lot of data right I mean this is actually much bigger than I normally do and it looks like for about a half hour video lovely rest of the

day bye-bye we're already at uh we're already returning like 900 so that's kind of annoying but anyway we obviously got the data now in my head I'm thinking okay like why don't I just go and I see whether or not I can turn this into just one big transcript it'll make it a lot easier for me to like feed this in a chat GPT so I'm just going to say transcript and then what I'm going to do is I'm just going to join all of these um hold on a second ah sorry we can't actually

do this because this is being output as multiple bundles right one two three yeah so actually what we have to Do first we have to aggregate this to an array because the um join feature is an array only function so I'm going to go down here to array aggregator drag that puppy in the source module I'm going to set as get data set items and then for aggregated Fields really you think about it I should just do everything CU who knows if I'm going to need access to some of this data later okay great and

then in set multiple variables what I'm actually Going to do is I'm going to um actually have to run this first to get the data so We're looping through a th and now we're going here we've turned all this into an array ideally with a TH or 95 entries cool so we've just converted this into something more workable now what we need to do is we just need to map this Array to extract only the thing that we want which is the text so I'm basically going to return a primitive array with only the text

variable here and so to show you guys what this looks like if I run this again we are now getting um this big array with the whole transcript and now that it looks like this all we need to do is we just need to join this so we Need to connect all of the array elements and just stick it into one big string and I'm just going to do so um by using a space delimiter so I'm going to click okay now I'm going to run it one more time and let's see the output it's

just going to be one huge transcript which is beautiful okay so now there are a couple things that I'm thinking about like when I you know and I'm not building the system linearly by the way right we know That we're going to have a couple steps before we're going to have a couple steps after this is just the core step so it's the very first thing that I want to do but if you think about I know that uh there's limited context windows for AI so context window GPT currently sure I've already searched this up

as you guys can see let's see what the context window is okay so context length of GPT 40 is 128,000 are these the maximum context Windows so um for context this stands for 128,000 tokens now tokens to words looks like in English a word is 1.3 tokens so 128 / 1.3 it's but 98,000 words can I actually get this in characters tokens to characters it's probably simpler okay one token is four characters in English that seems pretty simple one token is equal to three quarter words so one token is equal to four characters so sorry

so I guess I did the math So uh 28,000 * 4 about is about 500,000 so I just need to make sure that the inputs if you think about it are less than like ideally be way shorter than this but I need to make sure they're less than uh probably about half of this because I'd want the input to be half no actually no no no never mind probably about 300,000 400,000 I would say the longer the contacts sorry the longer the prompt that I'm providing obviously the more money it's going to cost me to

do This big content summarization stuff but uh that's sort of what what's running through my head right now like I don't want this to cost forever uh like a million billion dollars so what I want to do is I just want to get like um I want to get a feeling for how many characters this is so I can see so this is 34,000 characters so um considering that my kind of token limit here is 52 like this is that's crazy right 512,000 34 my video is half an hour and it's 34,000 characters how long

would a video really need to be if it were to make it to 512,000 probably be stupid long so actually if think about it I don't actually need to do any any funky special math here I'm just going to like feed this whole thing in even if it's twice as long and it's 70,000 characters I mean it's still like one 1/8 17th of the total context window of the model so it should be okay it should understand Everything that's being said and we should be able to use this to convert this into some sort of

content okay with that out of the way I'm going to go to open AI I'm going to create a completion here and while I'm at it why don't I just name some of these so this is going to be um con catenate into string this is going to be okay and I'm going to want to generate a bunch of things here if you think about it like I Have the opportunity here to generate not just uh not just uh like whether this is not not just a summary but I have the opportunity to generate like

four or five different outputs I could generate like a Twitter post I could generate like a like a LinkedIn post I could generate like a YouTube video script so I want to do this as in as intelligent a way as possible possible and so what I'm thinking of doing is right now I'm thinking I will Take this big long concatenated string that I've just gotten as the transcript of YouTube video and I'm going to turn that into an outline now this is a form of structured data that I can do a lot with once it's

in an outline I can then take this outline and then I could use it to recreate a post or to add new things to it speaking of adding new things I think this is a good opportunity as well for AI to suggest some new topics ideas that We can combine so why don't I just call this process transcript and then what I'm going to do here is I'm going to go into my gp4 module go to gp4 is 01 mini really fast I don't actually know which one's smarter maybe we'll try Owen mini for today

then I'm going to do messages and the first thing I'm going to do is I'm going to add a system prompt that says you are a helpful intelligent writing Assistant this is just how the model is going to identify so I'm saying hey your identity is a helpful intelligent writing assistant if you think about it logically this is sort of like the the the number one place to steer the model it's less important now but you know let's say you are um you know taking off at some I don't know this is like some big

land mass here and there's some Port over here okay and you are in a ship I should be able to make this text Smaller there you go and you know you are trying to get to some destination here and in between you guys a bunch of water you want to go over here if you think about it I'm just going to write D for short this is my destiny okay we're gonna we're gonna we're going to go here the angle of the direction that like the direction that I go at the very beginning is very

Important if I'm off even like one degree here I go in a completely different direction to if I'm here completely different direction than if I'm here completely different direction than if I'm here right this distance here might only be like one little degree but then the result is I end up let's say like 500 kilometers away from my goal so this basically the system prompt at the beginning of your prompt you have The highest likelihood of making a decision and steering the model in the direction that you want to go okay and this is um

sort of my my idea behind you want your model to be exactly what you know you you you want to provide these instructions and you want the model to identify as as smart uh uh a model as humanly possible so I say you're a helpful intelligent rting assistant because that's what I want I want it to be helpful I want it to be intelligent I'm sure there are better ways to optimize this you probably can write some super crazy long system prompt but this works for me in like 90% of cases so it's what I'm

rolling with um but that is just in a nutral explanation what the system prompt works I'm saying this because I've had a few comments and DMS and stuff being like hey you know what exactly does system prompt do the next is a user prompt so this is where we give it the instructions we should Really just call this the instruction prompt instead of user but anyway um we are generating content based on YouTube video transcripts so I generally just give it some instructions uh some context of my task then I say your task is to

take as input a full transcript of a YouTube video convert it and output variety of parameters let's say I'll say Summary um several paragraph summary of the key findings of the transcript outline a detailed chapter heading chapter style outline of the transcript written as if you were outlining a book um suggestions three three added three twists you three twists Or three uh let's do three pieces of added value three pieces of three ways we could modify the content to improve it discuss different ideas provide higher quality information Etc see and then social media posts hey

Twitter post uh Twitter care length car length 280 I think I think it's going to be 280 a LinkedIn post medium um let's Do 1,000 cars and a YouTube script WR let's just do 10,000 cars this looks reasonable I will say um that does not copy the original that includes similar content as the transcript but written differently let's just do that okay and then let me just see if there are rules we'll say write um I always do this because I don't like how AI tends to Just write a bunch of and it goes Super

flamboyantly and everything's extravagant and amazing and dive and delve I don't like that so I like to say right in a casual but Spartan tone of voice No Frills um output your response in Jason using this format and now we're going to create the Json format here so I'm going to go summary um outline and let's do markdown an ATX Markdown format um suggestions and we're going to do an array of strings and then social media posts we'll go post and then I'm going to have an object where it is is Twitter um linked in

and then YouTube script should be good let me just double check that this is actual good Jason the reason why I do this even though it Doesn't matter uh is because I want the model to be as smart as possible if it starts seeing some issues in the way that I'm giving it instructions on the Json um then it'll probably be a lot less likely to do so okay great this looks pretty solid to me um now I'm going to go to the prompts engineering again then go user one more time and I'm just going

to feed in the text right and I I just want to see what The output is um it's hard for you to obviously make uh suggestions or changes or anything like that if you don't know kind of like a core input last thing I'll do is I'll go to response format click Json object parse Json response very important otherwise you're not going to get the uh text as an output as adjacent and generally speaking just just some background on the prompt engineering side of things I just like to be as straight forward and no is

humanly possible um looks like the system prompt is not supported with ow and mini so I'm actually just going to go to GPT 40 I'm not sure what this means I don't really have the word withth all to fix it going go back to GPT 40 and then we'll click okay um but yeah I like to just be as plain and as No Frills as humanly possible with this Stuff uh and another rule that I always like to think of is just try and get something on the page as quickly as humanly possible because if

you don't like if you make all of these assumptions in your head about whether it can up with the transcript and the summary and blah blah blah um you're usually going to end up being disappointed so I'd rather just like know what the what the actual result is okay so the Json format looks good Summary outline suggestions post that's nice summary cool outline I'm going to need a much more detailed outline than this if I'm going to want to generate content with this later I'm going to want more detailed for sure suggestions I don't like

these suggest suggestions what I want is I want suggestions on the content I don't want downloadable checklists or real life Studies well actually I guess that's not that bad this is technically still content suggestion expand on the theory with specific example successful abols guess we could do that my idea is I'm going to feed in the outline and the suggestions and then I'm going to to another prompt and we're going to have it like create some better content okay uh I don't like the Twitter posts YouTube post is all right I guess sorry The um

LinkedIn post is all right this is obviously a little Limited in terms of length but H yeah I don't I don't really know how much value there is in generating a YouTube script to be honest but I just wanted to show you guys that you gener multiple different types of content okay well let's tackle this one at a time first of all the summary I don't like that it's as in this video Blah Blah Blah Nick outlines blah blah blah um I would prefer it just to say I would prefer it to write well actually

I think i' prefer it to write in the first person than the third person so I'm I've outlined a process for building effective no-brainer offers and sharing insights that help my agency reach this much money like if you're going to be you know doing this content yourself you don't need this level of summary then The outline I want it to be a lot more detailed like I I like this but I want the um like the subheadings the uh bullet points to be substantially more detailed than this the suggestions yeah on second thought these seem

fine and then the posts here with the Twitter and the LinkedIn I don't really like the way that they're written so let's just go through these one at a time we're going to try providing some more Instructions okay I'm going to say write in a firstperson POV or use a firstperson POV for everything you are the Creator [Music] um an ATX markdown format go into extreme levels of detail in your bullet points you should be able to recreate the you should be able to recreate the content from this uh from the Outline most of the

content using solely the outline let's try this suggestions three ways we could modify the content to improve it twists different ideas uh areas we could add more context Etc social media post a Twitter post a LinkedIn post post include Sim write these do not sell don't add call to actions in these posts cool this looks reasonable to me Um what I'm also going to do is I'm going to compare the quality of GPT 4 to um to 01 as well I'd be very curious to see um whether 01 can do better because this is like

multi-step prompt there's a lot going on here so if we can have it like output three or four um Chain of Thought steps in between I think it would do much better I'm going to change the temperature to 0.8 as well Max completion tokens I'm going to set to can we do 8,192 let's see let's try this one more time now I'm about to sneeze nope ah nope got to got to handle on it that is strength right there that is some willpower ah okay now we have outputed everything written in first person I think

the outline should still have more detail I think like definition of high leverage and automation X Y and Z like basically we should say what it is but uh this is much much better than before and this is great this summary is awesome let's just just see what the suggestions look like Implement more real life case studies explain more about automation tools and how they can be leveraged effectively discuss common challenges face and creating Nob brainer offers and how to overcome them I'm going to want this to be a little bit more unique maybe even

spicy I'm not Sure maybe if I use the word spicy this will be better um H twist be unique be extremely unique and spicy and what was my other issue with this the outline don't just say for instance don't just say leverage I discuss leverage say I discuss leverage the concept of multiplying your output through intelligently Um simple machines like automations can multiply your output through I've constructed intelligently cool all right all right we're going to go okay here I'm going to click run once and we are going to look at this and then I'm

going to cross compare this now with um the output from um gpt1 as well and I'm going to basically just see which one I like more and whatever one I like more is just the one I'm going to move forward With cool summary no changes there outline cool o this is great nice yeah yeah okay so this has done basically exactly what I wanted right so every heading now has like explanations of basically what what they mean what look it even gives me like a little formula for the risk minimization that's great suggestions introduce a

segment where i deconstruct a failed offer now This is great incorporate customer testimonials or case studies yeah fine add a live demonstration setting up a new Briner off from scratch look at this this is sweet okay awesome awesome this is great this is all right I mean the Twitter post is not perfect but whatever you can obviously spend as much time and energy as you want um flushing flushing all this stuff out cool so now that we have this Transcript I'm just going to build out the rest of the kind of the middle part of

the logic which is um we need to send this and put this somewhere right so uh I'm just going to go to Twitter X form and then um let's go create a post oh I'm going to have to create a connection here that's pretty annoying I don't remember how I did this before give me one second to think about this deeply okay depth of thought completed Um why don't we go back here jump over to Twitter and what we need to do is we need an X developer account let's go over over here set this

puppy up cool then we need to create a custom application INX developer portal sound for a free count describe your make use cases select checkboxes submit um I feel like I've already done that right projects main I think I did This forever ago so why don't I actually just like rebuild all of this stuff for you guys how do I delete this am I blind these are my keys and tokens right but I don't actually want this I just want to delete it so you guys could see how to put this together H fascinating I

don't actually remember maybe I need to edit it ah there we go Delete app cool example app Aug 12 104 okay so now we're going to go here I'm going to click add app I'm going to say um I don't know tweet poster no uh let's just make this really unique okay now I have my API key API key secret bear token so I just wanted to double check what do I actually need to make this happen client ID and client secret so I don't think this is my Client ID and client secret but I'm

just going to copy this and why don't I just put this here in a set multiple variables module let's just go API key and then I'll go API key secret and then also go Bearer token just in case I need it later I just want to have it cached okay and then app settings then user authentication so setup I want read and write and direct Message I want uh automated apps and Bots I think app info callback URL H this is interesting um I think I need like the integrat Callback URL so let me just

go to these instructions here so I'm just going to copy the Callback URL paste this here and this so I need two actually and then I'm going to add this paste this in then I'll go back here and then for website URL copy this paste this In the rest of the stuff looks pretty good so I'm going to click yes okay now I have the clan Ting Clan okay great so now I'm going to go back to this module create a connection paste in the client ID paste in the client secret click save probably going

to have to sign into Twitter now I imagine yep that's fine don't mind if I do do I remember what my Twitter password is hold on a Sec Twitter Nick Sur okay yeah that's definitely that uh this going to be a tough one guys ah sweet I did actually in fact remember B aut full and now we have the connection on to Twitter lovely I feel like this is such a common question actually I should probably like make a video just on this anyway then we paste the text uh text content of Twitter there's some

advanced settings here I wonder what these are media ID so You can actually upload media and then link it Place ID direct message link oh that's cool a lot of post exclusively for super followers only quot post all right we don't need any of that stuff so uh then I'm going to say create X poost I'm we go to LinkedIn create a user text post here and yeah we do need a router if we're going to do this I'm already uh connected but the authentication is a lot easier you just click add and then you

connect to your Account the post in this case will be LinkedIn uh and then you know what I also want to upload my script somewhere so I'm going to go Google Docs I'll go create a document and I'm going to want to upload this YouTube script let's call it YouTube script what I need so I want how to create no-brainer offer which is going to be the first Element the content of this is just going to be this here um did I write this as text yes I did okay great and then the document location

I'm just going to do like the index here that looks pretty solid so now I have the X post being created then I have the LinkedIn post being created then I have the Google doc being created all good stuff um this Google doc is going to contain The YouTube script right and I'm doing this just to show you guys that you can you can output basically whatever the hell you want here if you wanted to make this higher quality content just for LinkedIn or just for uh Twitter or something then you know spend spend a

little bit of additional time uh doing so and then why don't we why don't I also show you guys you know if I were to build out the system why don't why don't I show you how to do the um suggestions Improving the quality of the outline as well yeah let's do that so actually just before I do this why don't I go over here and add an additional prompt and I will say upgrade transcript maybe upgrade outline upgrade content upgrade outline um and now that we have GPT 40 here your helpful intelligent rting assistant

we're generating high quality content based on YouTube video transcripts your task is take us input a An outline for a YouTube video and then implement the provided suggestions to make it significantly more detailed okay great rules WR a casual but bur tone of voice no for use a first- person POV for everything you are the Creator I don't think we need this anymore output your response and Jason you'll be provided a list of three Suggestions and the outline modify the outline leaving as much of it the same as you can no modify the outline and

add detail output your response in Jason using this format we'll go modified outline keep the outline in markdown etx format modified outline will be here and then uh yeah I should be good I'm going to go and now feed in the suggestions and then I just have to Reproduce this unfortunately so I'm going to feed in actually let's do the let's do the outline first then after that I'm going to go suggestions it's an array with three elements the first element is this one the second element is this two and the third element is this

three and then we're going to keep the output as Json pars Jason response Max Tokens fine whatever um all right sweet okay great and yeah I'm realizing I'm doing this sort of backwards because now I'm upgrading the outline but I just wanted to show you guys what this looks like in practice first we have to process the transcript get the suggestions then we have to upgrade the outline right this is sort of like a more Dynamic kind of uh loop we are getting it to create suggestions on the content then feeding those suggestions Back in

alongside the content you can do this arbitrarily as many times as you want just to like get new twists on the content cool wow this is much more detailed now nice oh this is great sweet so now we also have like a core piece of content that we could like yeah leading into upsells example offers we we have a core piece of content now that we can actually like modify and and do other stuff with like if you think About it what we could do is after we process the transcript and then upgrade theine we

could actually use another gpg call and use that to create new content for x and um LinkedIn and Google Docs as well I guess in this case I was just I was just trying to show you guys how to like generate said text but um if we're going to be going all this way then we should totally upgrade all of us so actually what I should do here is I should now generate content and then What I'm actually going to do is this first step here I'm just going to just going to make it a

lot simpler I'm not going to feed in the um social media posts I'm going to actually generate the social media posts based off of this so I'm get rid of that we're just going to get rid of the post object completely it's going to make it a lot simpler um and I think the summary is still useful so I'm going to keep it we might need it at some point In the future and I don't know cost very minimally cool then upgrade the outline then we're going to generate social media posts beautiful and then I'm

now going to feed this instruction in we are generating high quality content based on content for social media platforms based on a provided outline I will say your task is take as Input an outline and then I actually let's just paste this in and then produce social media posts for Twitter LinkedIn uh why don't we just do more Twitter Facebook LinkedIn YouTube uh what else do we got really what are the other major social media platforms Instagram I guess Twitter Instagram Facebook LinkedIn and Linkedin as well as a YouTube Script then for rules I'll go

and I'm actually just have chat gbt help me with this I just held option press space and I'm using chat GPT which is a desktop application on Mac OS um I'm adding uh specs and guidelines for social media generation post generation I know Twitter LinkedIn and YouTube modify the below post to include Facebook Facebook and Instagram as Well I'm not doing this because I necessarily would have um like in practice I just wanted to show you guys how I tend to use large language models in AI in my own work so I basically I mean

you know I'm writing a prompt for an AI using an AI now right okay uh no refs cool I'll put your response in Jason and we'll go let me see here Twitter Instagram LinkedIn Facebook YouTube oh I shouldn't be capitalizing these either you generally don't want to capitalize the first character of a variable YouTube script I think this should be good anything I'm missing LinkedIn YouTube Facebook Instagram Twitter no I think that's fine medium visually engaging designed a pair no we're Okay okay YouTube script contain uh including a YouTube script that breaks down the outline

in detail and then I'll say write in the first person POV you are the Creator okay awesome so now we're actually going to generate these social media posts we're going to feed those in and we're chaining three prompts together here right so the first AI model does something to the transcript the second One grabs the outline of the thing that we just did uses it to upgrade the outline and then the third um takes the outline and use it to generate a bunch of high quality social media posts presumably high quality social media posts that

are now now pre-drafted for us to go in there and do stuff with um yeah so now I need to basically fles like the database logic and stuff and I'm I'm not actually going to leave the posting part uh for the very end Just to show you guys you could automatically post this if you wanted to um but I'm probably not going to I'm just going to create a big database to store all the stuff and then whoever wants to can do whatever they want with it okay and looks like the result is out cool

this is pretty I don't know how I feel about the star LinkedIn Facebook cool and then Long YouTube post Okay ideally we'd have way way longer of a YouTube script so H um you know what I could do if I really wanted like a super long YouTube script I could go through the outline chapter by chapter then at the very end I could feed in I think I should do that I think I should split this based off of headings and then I should feed in the whole thing and just tell to like remove any

Repetition yeah that's probably what I'm going to do okay let's just make this way better I mean because right now the script is kind of trash right so I'm going to take this script what I'm going to do is I'm going to take this outline sorry and then I'm going to um separate it based off the presence of these so these two number signs I'm going to have it write like a very detailed section for each so one for introduction one for high leverage key Concept one for understanding the automation hisory one for crafting no

brainer offer leveraging templates examples blah um I'm going to feed it in the summary of the post as well just so it kind of gets the idea and then after it's done with that I'm going to pass the entire thing through GPT for and just say hey you know if there's any repetition or anything that sounds weird here because You're repeating the same thing from one section to the next then let me know but I don't think there will be let's do that instead of this the YouTube script part is just not not high quality

at all so I'm just going to add one more GPT for prompt and um yeah we're not going to do the YouTube script uh let's remove this too let's generate generate YouTube Script what we are going to do is we're going to have to um um going to have to go to the iterator sorry using the wrong one here what we're going to feed in is we're going to split um this outline modified outline based off the presence of uh this and then a space I believe this is going to turn this into a big

array we're building an no-brainer Offer and then we'll have an next section the next section the next section um and then I'm actually just going to cache this I'm going to say outline I'm actually just going to take as output the input of this section here because I just want to run a bunch of tests on this very quickly I don't want to have to use God knows how Many um operations in order to do so so now I'm just going to unlink this drag this puppy all the way over here and then we're going

to get your ass over there and then uh I'm going to feed in the outline test variable instead let's see what happens that's okay so now we are iterating through this is the first bundle second bundle third bundle fourth bundle fifth bundle sixth 7th 8th 9th cool so now we Have all of the text looks like we're just going to ignore that first one so I'm going to feed it into the generat a YouTube script Modo now like line by line and then what I'm going to do is I'm going to set thing where it

says bundle order position not equal to one I don't want that first bundle to get in I just want all the other ones so now we're going to generate YouTube script um procedurally like Chapter by chapter we're generating high quality content for social media platforms uh we're generating high quality YouTube script script for YouTube upcoming YouTube video titled um what is the upcoming YouTube video titled I guess I don't have access to the rest of this let's just stick that here so I Could finish this prompt hm I need to click okay here to refresh

the variables this ever happens to you you can just click okay and uh give it a refresh what is it going to be titled um let's go as early as possible there you go I'm going to say we already have we are doing this section by section your current section is and then I'm going to feed in the Section which is this value a summary of the entire video so you have context is and then I'm going to feed in the summary variable here your task is to take as input this section your task is

to is to produce an engaging script section piece of script for this section let's then Do um output in markdown ATX format using H2S and h3s your first um heading should be and then we'll go the title of the section do not add ctas uh section should be extremely indepth and written in a casual but Spartan tone of voice no FRS right in the first person and then let's just remove this and go output your response in Jason using this format we'll go generated section cool um and then after we're going to feed it through

so we're going to provide actually we don't actually need to provide anything yeah cuz we already provided the the text up here so that looks Pretty good um we do need to C you know we need to cash the summary unfortunately because um we're getting the summary from earlier so I'm just going to save this go back here and just see like an earlier run and it doesn't look like we have a single run here really come on man oh yeah we do this just lied to me we'll go back here to um one actual

execute ution I'm just going to grab the summary from this and just paste it in manually Summary here we'll go back to the diagram and then I'll just use this as a test sample uh because if I didn't then I just have to jump in there and change it all the time sorry did I really not save my work you're serious come on man seriously I need to re I need to redo all this no uh can I process a previous version 1015 I think I think that's the oh no no no no no that

sucks 10 10 oh That's terrible oh I hope to God this has some incling of my prompt no no it does not okay well that's annoying I'm gonna have to rewrite this I'm not going to include that part in the video so I'm going to cut it out just for time purposes make.com guys not even once let's go to uh what was the system called YouTube scraping system Okay history let's go to this fabled run the one that screwed everything up um process transcript and what I'm sucking out of that is the summary cool I'm

going to go back here close this and voila I still have everything in place thank you thank you let's just do some standardized formatting here too okay this looks good um we are going to grab the outline we're going to iterate through it we're going to Generate a bunch of these and then once I'm done generating a bunch of these I'm going to uh contate them all back together with another set let's do set multiple variables I don't why I don't there's no need uh no reason to just use the set variable module really you

don't waste any Ops and it just allows you to recreate stuff anytime you want then I'll go concatenated um script and then I'm Just going to join oh sorry actually I need to use an aggregator now we'll go aggregator again I'm just going to get the multiple bundles and I'm just going to stick them into one so the source is going to be the iterator this is the thing that makes the multiple bundles and then it just needs to be equivalent I'm just going to select all the fields and then here what I'm going to

do is I'm going to Jooin the output and unfortunately we don't have access to it because we um need to fix this first to join um oh sorry we need to map this array we need to get result but it's not enough for it to be the result we need array. result. uh we need something called result generated script I think let's just see how did I word this in order to map you need to um use the actual key name not Just what it thinks the key name is just delad this because we definitely

don't need that uh result. generated section okay result. generated section then we need to join that and I'm just going to join it with a new line okay okay and let's go back here ensure everything is fine I think so click okay then let's click run oh what's going on here we don't have the schedule in the right place there you Go uh hold on I'm still referencing something from a previous module which is a problem so I'm just going to make sure that I don't actually need this where am I referencing this oh right

there we go okay we iterated through cool cool all I'm not generating the text generating the script I should say much higher quality script than Before like you know I figure guys if you're going to do this you might as well do it right um the the value of doing it with this approach is GPT will use usually have like a soft character or token limit output so if you try and get it to Output really long content you're like I want it to be really long it might only generate you 500 Words and you're

like I want it to be longer and it'll generate you like 100 extra ones I want it to be Really long it'll only get you like 50 so this way we're doing so we we're basically just generating section by section we're going pretty detailed and then I'm just concatenating all that stuff together at the end um after all of this is said and done we're just going to have like a big um a big chunk of text essentially and then with that big chunk of text we'll uh yeah we'll we'll like basically a full script

so if somebody wanted to teleprompter this They could um what I'm going to do is I'm going to take the script the X poost the LinkedIn post the Facebook post I'm just going to dump it into a uh a spreadsheet so I'm going to go here to the Google Sheets and then I'm just going to make my own and we're just going to call this content database let's say YouTube scraped content database you think about it logically We're going to need uh like a Twitter post going to need like a Facebook post we're going to

need like a LinkedIn post we're going to need like a YouTube script I think we should just store all of these as their own Google docs to be honest rather than just link the plane text here because there might be some formatting or something or or I don't know sharability reasons it's probably what I'll do uh what else did we have we have Facebook oh Instagram Right let's do this Instagram Twitter Facebook LinkedIn okay cool and while I was doing that um you know I did that cuz I wanted to make sure this could work

now we have a very long script right much much longer so why don't I just take the whole length of this really quickly and then I'm just going to feed this into word counter and just see what the length of this is 37,000 characters so what I what I'm Now going to do I'm going to do one final GPT for call I know it's quite a lot but I'm going to do one final one and I'm just going to say remove any inconsistencies with the script like hey you know are there areas where I'm repeating

myself because sometimes the model the model has access to very isolated information here right it can only see the section it should write plus a summary so it might repeat itself a couple of times because it doesn't Really know the context of the other ones so I'm just going to say smooth out script I'm just going to call it that and then here I'm going to write um another prompt The Prompt your helpful intelligent writing assistant we're gener a high quality script for an upcoming YouTube title video title blank um I'll say YouTube video I

will say however our script was generated section by Section and so each section lacks context on the sections next uh lacks context lacks the context of The Wider script your task is to take as input the script script and smooth out any inconsistencies repetitions or unnecessary sections you then you will then output the response in markdown ATX format cool now Um leave all this out I'll put your response in Jason let's say formatted script and then I'm going to feed in my input my input is going to be this giant ass concatenated script script and

say um poorly or maybe disjointed script when don't we do that might be a little bit better and then we'll say smooth script uh clean script I don't want to like um give it a word that implies that it should chop it Down a lot final script let's go rough draft script cool uh oh maximum completion tokens I just realized I don't know if we'll actually be able to do this um H GPT 40 output length Max 4 8,000 o output token limit 496 16 384 oh man that's not probably not going to work well

wait wait a second wait a Second what was the word counter 37,000 so 37,6 191 divided by 4 9,197 okay actually you know what this is probably going to work on things up to about an hour long uh yeah never mind I take that back thank goodness that would have sucked okay uh let's actually run this now and I don't want to have to do all this again so what I'm actually going to do for my test is I'm just going to copy This whole thing in yeah let's not actually run it on real data

let's run it on cache data that way I just don't have to wait and consume 8 trillion tokens right okay rough draft script boom okay let's go here drag this in and see the final product and then I'm just going to read it really quickly Ensure that it's like an okay script it's not total hopefully it's not total trash fact that it's being output as Jason I think it'll be all right uh whoa no no that is not good this is very very very short H that's not good at all we don't want it to

make any changes like this we wanted to just um rules keep the Script the same only change something if it's unnecessary or repeated the length each section should look very similar your output should look very similar to your input with minor changes very minor changes um I'm going to feed it the input in not as rough draft script go input Script and we'll just go output script maybe this will help oh you know what actually sorry it's probably my token limit I set 8,192 token limit right um did I a Max completion tokens here we

go maximum number of tokens to use low values may cause it to be truncated High values may use a lot of open AI credit or reasoning models such as the own models Val s reasoning up tokens uh 16384 is that I think so maybe explicitly saying 16384 will help like if you think about it my my input here on word counter was 5,991 91 words so I want something that's like probably 4,000 words or higher I don't want anything shorter than that otherwise probably illustrates that there was a major problem here if I put 4,000

words um 4,000 words would be approximately equivalent to like 7,000 tokens or Something that's fine and this will I you know as I mentioned work on anything up to an hour the fact that it's taking a really long time to generate is a really positive signal for us it's telling us that like odds are it's generating a bunch of tokens right now which is great crossing my fingers I like what I'm hearing or not it's funny because I you know I wanted this video to start off as hey You can very easily scrape appify and

then I put a bunch of sections and now halfway through this video I'm like wow we could totally build like the most epic content generator with very similar flow and I'm thinking hm you know I should probably update my content generator to be a little bit simpler and to employ strategies like this um as opposed to what I was doing before which I was doing some very interesting like outline based prompting with arrows and Stuff okay so now we're at that point now where it's getting so long that I'm starting to think that uh something

weird is happening I might be running into some upper token limit or I don't know might be rate limited or maybe it's broken who knows but that's taken a little too long definitely not a fan of how long this puppy's taken okay result output script much better much better okay I'm Just going to copy this I'll go to my word counter the input was 5,991 words okay the output was 5,989 words wow we actually added some interesting I'm going to be very clear about what the changes were I'm going to go to diff checker.com paste

in the changed text here I'm going to paste in my original text here and I'm going to find the differences so it's removed a bunch of the same introduction sections which is Good why don't I just look at the removals removals or where that so removals this section here H what is the difference here one lesson is it over promising under delivering I'm not seeing any any changes here at all oh or or sorry is it saying that they removed this in the yeah sorry I think they're saying they removed this in the output or

something or maybe this did not have this this had this and this Didn't are they just saying that they've added space H strange yeah I'm seeing seeing very little change here yeah so this is this is the stuff that I don't want welcome to this video right so that's not good we don't want this um doesn't look like my Approach necessarily is worked unfortunately how can we fix this Thinking out loud here we could we could try a different model so while I'm doing all of this why don't I try a different model let's try

one of those o1 models and see if we can um you know just leverage smarter or more intelligence basically I think I want this one the preview and I'm going to run sorry there's no system prompt in the o1 I forgot about that so I'm going To go back here just really quickly remove the system prompt click run once again temperature does not support 0.8 okay all right all right all right we got to go change a couple of other settings here cuz this model only has a response format type Json object is not supported

with this model okay okay okay we'll go response format text let's just go empty actually Uh what else what else let's just remove this Jason part okay okay we got the result result is still pretty long but it does look like we were truncated we were cut off why were we truncated or cut off I'm going to go um can we go back I'm going to feed in the output then I'll also feed in the input oh I fed it in his Jason too which Might have screwed things up a little bit okay and then

let's find the differences oh now this has a ton of differences oh this is great wow oh I get it it added a whole new section at the beginning wow this is way more intelligent wow oh man it's even using the term leverage has some minor changes here yeah yeah yeah yeah wow wow wow this is great it's like Readjusted and mooved things around and it's removing a bunch of restatements of the same thing just completely removing things that we've like duplicated which is what I wanted cool okay well um listen this is fantastic the

only issue is obviously this ends a little faster right well actually I guess it just like removed it completely we are we are really close to The ending I don't know why it did that is this a token output issue it's paste in the 5,991 sorry that's the intro let's um let's copy the what am I copying here am I copying both of these things the up put's 5,8 words 3,17 oh this is actually really close to the character limit 3,176 9id by 4 it's about 8,192 so you know what I'm thinking it's probably just

caps out at 8,192 damn this is good Though this is what I want yeah this is like an actual script that I'm thinking I could probably use on my own channel like that's how that's how high quality it is so like I don't use Scripts or whatever obviously cuz I've just been yapping on and mumbling but um you could totally do something like this well I guess what we need to do in order to make this work then is we just need to make sure the output script is almost always less than uh 8,192 tokens

To be honest so let me check 01 preview output token length no weird 120,000 tokens for both ow one preview own mini 32,768 output tokens and 65,000 for ow and mini I bet you if I used ow and mini it would be stupider though right I mean it has to be stupider I'd imagine it's stupider man how the hell are we going To solve that that's beautiful that's exactly what I wanted uh could I the output length a double maybe it's just because I set the output length wrong maybe we'll do that yeah let's just

try this one more time um while that's generating and taking sweet ass time let's just finish this database feature okay so in order to do this I'm Just going to go back um because uh make does not allow you to click into these other modules while we're running um an execution I'm just going to go back to a previous execution I'm just going to pull all the data of that previous execution out the idea being um I I need to get a bunch of headers of the data that I want so I'm going to go

to diagram and then I'm going to go way back probably like Here to see this get data set items appify module and it looks like we are outputting this information here video description video keyword so I'm just going to want to copy this go here and say generate a list generate nope generate a list of CSV headings based on the key names here copy and pasting into sheets paste this in it's now going to generate me a list when this is done I'll be able to Copy this over to here why don't we do this

uh hold on actually let's just take this and paste this down here let's go over here and then go um data split text to columns then I'll go here and then paste this in then I'll make these the headers of my column of my thing we have Instagram post Twitter post Facebook post LinkedIn post YouTube script YouTube script is the big boy that's what I'm like you know obviously it's The thing I care about the most because uh I've had a lot of fun putting this together so far so I'm going to go enter and

then just sort it like this then basically what I need to do is I need to I need to create four Google Docs one with the Instagram Twitter Facebook LinkedIn YouTube script and then I just need to add the Google Doc links right here which is going to be straightforward I don't need the text Column now that I'm thinking about it I don't need the duration column I don't need the start column either author yeah video keywords video description video length video title video URL video ID awesome so everything else looks fine cool and while

I was doing that this just finished operating and it finished with the whole script oh incredible yeah yeah yeah it made it it Made it oh nice so you know must have been an issue with the output length maybe because it goes multiple times it calculates it differently I don't know but this is this is great this is exactly what I wanted that's so cool oh man it's moments like this that like I get genuinely impressed again with these models it's been way too long okay uh H I'm going to now let's see here copy

and paste this Behemoth and I'm going to add a sleep Module over here because I know this is going to consume a ton of my tokens so we're going to sleep for let's say 5 seconds after we generate each YouTube script section even on the last one it'll sleep for 5 seconds then it'll feed it into open AI beautiful and then once we're done with this um we're just going to create Google Docs basically I could create Four I don't think I'm going to do that I'm G to use a router like I will create

for but I'm just going to split it so we're going to call this um Instagram post actually let's feed in the title instead title's probably best if you in Title First Instagram the content of this output going to be the Instagram and then uh I guess we should Make a Google drive folder first actually so after this is done let's create a folder this folder is going to be generated here I'm just going to generate a Google drive right now called YouTube content scraper I'm going to feed in this as the location of the folder

we're going to call this the title of the Video and then we are going to share this folder the role will be reader um yeah screw it let's just share uh no we want a we want a writer anyone and then I'm going to now generate a Google doc the Google Doc will be specifically inside of this location it's a new Doc location will be here and then because I already have the title in the Google drive folder I don't need this we'll just say Instagram okay and then I'm going to do the same thing

for Twitter and then um LinkedIn and then Facebook and then also my YouTube script so 1 2 3 4 I'm just going to Auto line these make them prettier and then I'm just going to go Twitter Facebook oops no need for me to clone that my bad then uh LinkedIn and then YouTube Script that's G to be the big boy awesome um now that we have all that all we need to do is just go in here and then change the variables so this one was Instagram I'm gonna go Twitter it's going to be Twitter

this one is going to be uh Facebook this one is going to be linked in this one's going to be oh you know what we can't actually do that with the YouTube because right before we do we have to um convert this into Markdown so I'm going to go sorry we have to convert this from markdown into HTML so HTML to markdown there we go sorry I mark down to HTML is what I wanted markdown to HTML can delete this then Mark down HTML I'm going to feed in the markdown of my mega post here

beautiful I'm going get the HTML out of that and then this is what I'm going to feed in as the Text and I'm realizing that I didn't change these titles and this was Twitter okay now that most of this is done we should have everything we need in order to actually run this on live real content before this I'm just going to feed in one more sleep module in between each of these for five seconds because just given the sheer level of token usage I have a Feeling that I'm going to be running into some

upper limits here um and then I need to run through and just change some of these variables so where it says generate YouTube script if you guys remember I had I was using the wrong variable um because I was just like using the same title so I'm going to go back here to my title then I'm going to feed this in we are doing this section by section Our current section is value summary of the video is here so I'm going to delete all of this and then put in the actual summary of the video

which was right here um that looks good and then smooth out script here I just fed in the whole script right so we can't actually do that we actually have to go all the way down then I'm just going to feed in the whole input Script where did I get the input script um I believe it was right here right outline test variable or concatenated script yes there we go now this will be on live data and then am I using anything else that I don't have to be outline test variable it's feeding in here

this outline test variable needs to be the modified outline I think this is good I believe if it's not then it's what it Is okay I'm going oh you know what no no no no no no no hold on do not no Force the stop okay one more thing we need to do we need to add all of these posts right logically um we need to add all these posts and I think I just realized I'm not going to have access to these if I do it this way that's so annoying yeah sorry we can't

actually use this routing we actually just have to do this in a straight line Why because I need access to these links if I'm to use this to update a spreadsheet so one two three even though it's not really as visually pretty four we could also run um like an um an incrementer function that just like goes through each of these and then for one it goes Instagram for two it goes another thing for three it goes another thing but uh I don't really want To do that okay after we generate the YouTube script doc

now we're just going to go to Sheets we're going to go add a row now I just need to connect this to this Google sheet here so I'm just going to copy in the ID and I'm just going to enter manually because I know what all this is Go sheet one the column range is going to be A to Z this is now going to have all of my values um every last one of my values and then if you think about it what was A we go to my Google sheet here it was video

ID so I should just have all these variables like one for one here so I'm going to go video ID video URL title length description then video URL title length description then I believe we had keywords author and then Instagram Twitter faceb yeah keywords and then author so go one more time I'm going to have to scroll the way down here go keywords because it's an array I'm just Going to join these keywords with a space and then this last one here is going to be author looks like it's yeah then after this we just

pull in all the data here so Instagram Webb link um Twitter Webb link actually can just search up Webb link here make my life easier Facebook LinkedIn and then YouTube for column L awesome just check this this is column L right yes it is And I Think We're Off to the Races here I think we actually have everything I don't know if this is going to work but we have to run an an test to find out right so let's give it a go let us see what Horrors our work has created cool cool and

uh yeah so that's kind of that middle section there um the last thing that we need to do logically is we need to take uh we basically need a method that pumps in a bunch of search URLs here so outputs you know B basically outputs a a video transcript so there are a variety of ways we can do this I think what we should probably do is is we should probably do the run inside of YouTube Sorry inside of appify and then by doing it inside of apify um we'll just watch the actor runs here

whenever the actor run is completed then we will trigger the rest of this I think for Simplicity sake I believe we should Just run this once so we should run that YouTube transcript stuff just just a single time this finished really fast so I'm kind of hesitant here I don't know if this actually did my thing it didn't know we're missing something here unfortunately the messages that I fed were just the introduction that's all why we only generated one I don't understand that would mean that in the outline we oh oh it just use single

markdown huh uh H let's think here we split this based off the space right um okay you know where it says modified outline upgrade outline we need to just like give it some clear directions we just need to say output each chapter heading as an H2 cool there you go now we can run this puppy again um but Anyway you know we did this with like the shitty short stuff so let's just go through and make sure that everything else worked even if the actual AI part didn't uh okay I'm just going to make all

of these a little bit bigger so looks like we got all the data all the fields are filled yeah no problems there actually cool I am just going to reset the colors on this um because when you deal with uh make.com and Google Sheets it'll just And you add a row it'll just copy the formatting on the previous one and now that we fixed that I think this should probably be be better so we're going to run it again and then basically what I'm thinking of is uh you know this run right here right YouTube

video caption scraper I think we should just feed it in one at a time it probably be much simpler than if we do multiple so what what we should do is inside of make there's a way to Watch to see if an actor run is completed so if it has completed we'll then trigger the rest of our flow we'll grab the default data set ID excuse me and then we will use that to trigger this beautiful lovely sequence that I put together painstakingly and then after that um we will if you wanted to trigger this

on multiple what we' do is we'd have a Google sheet the Google sheet would have the video URLs the videos you want Processed and then that database would run when you want it to so you trigger the Run manually I imagine so you just click Start scenario and then it'll go through and then trigger this flow for all of the fut all of the rest of them and then in that way um I think this would be the simplest to manage and then we would just update and say processed at the very end I think

that's the way that we're going to do it so for instance this is Sheet one here this deals with the the actual scrape content we just go sheet two and then what we do is we go like YouTube video link we go YouTube URL and then status okay so me just copy these two because these already have the formatting that I want we'll just say video URL and then we'll say processing status when something is added to the sheet you will say Um awaiting when it is done it will say processed this will be the

Q and then we need another scenario and this other scenario is going to be responsible for watching this and then running it so we'll go let's make this one number one call this YouTube scraping system um run Scraper this will go to our Google sheet search rows it will search rows where uh wrong one it'll search rows where status is equal to processed so we'll go spreadsheet ID and then I'll go YouTube scraped content database sheet name will be sheet two for now you know you can make this whatever you want processing status must be

equal to awaiting okay then we're only going to Return URLs of YouTube videos like this one let's say that haven't been processed this will then trigger an appify module which will run an actor this has a red little circle next to it just because um I don't really know to be honest this will now run set actor we're going to do this um using I believe it was called the caption scraper YouTube subtitles was that what It was called let's double check genial candle stand yeah so we go genial candle stand integr will wait until

the actor run is finished uh maximum time for the run is 20 seconds no I don't really want to wait input Json we need to grab the Json form of this so I'm just going to go Jason copy that move back here and then paste this in what I want start URLs is I want a single URL and I want it to be the video URL there we go and you can set this to run on 15minute intervals you can set this to run whenever the hell you want oh geez there was some issue why

where did this issue come from execution was request did I you know what I think by me modifying the title I stopped the freaking run oh that kills me okay we're gonna have to wait a while longer um anyway YouTube scraping system run scraper this will be Watch runs and generate content there and then I'm just going to go over here and I'll go watch actor runs this is what's going to feed into the module and this is a web hook based event which will watch for the actor to finish um the run I believe

this is the most efficient way to do this I think that there's no more efficient way when you want to add something you fill Out a form maybe like a type form or something and then it adds a new entry to this with youtube.com whatever awaiting then once it finishes running you know this this triggers the appify Run once this finishes running we then run through this whole sequence and at the very very end what we do is we we add a row to the first thing with all the information but the second one which

is our Q um um we'll search for a YouTube video URL equal to the initial YouTube video URL and then we'll go to the processing status and then um switch the status from a waiting to processed so that way we have a Q here in sheet two which really should be sheet one if you think about it but we did this backwards um that you add to and then you have the actual like content generation side of the database which actually like you know fills it in and stuff like that yeah to me this is

the Simplest way to do it I think this is probably what I am going to do for time Reasons I'm just going to um stop the the Run well I'm not going to stop the run I'm going stop the video and then I'll just um double back after we're done okay looks like we are now wrapping up the generation we smooth out the script generating all the documents here looks pretty solid I'm liking this oo very cool Nice anything in a break you know it's very rare that my scenarios work on the first run like

this good god oh boy I'm kind of concerned now um let's just remove this first entry if you guys remember that first one was just like um from a previous round that was broken just open up our little content it's going to do some all tabbing back here Instagram content Twitter content Facebook content LinkedIn Content and the freaking YouTube script which is 10 Pages not bad not bad at all now you can do a lot with this a ton of things um you could take this content and then you could actually like write a whole

chapter with this as you as you guys could see this is actually really detailed um you could automatically I don't know you could like completely a generate the stuff if you want and actually the approach that I'm doing here like you can use this to run faceless YouTube channels to be clear um I think a lot of people probably are they generate things section by section by section they have some scripts smoothed out like that average word per minute um right now is about 150 or so so this is like 25 minutes of just like

constant speaking with a uh with an llm you could easily generate like very in detail 25 Minute videos on it um but you know for the purpose of this I'm just assuming that like you're actually part of some some content agency or something and we're generating these posts here so that the people on the team just have a rough draft that they can go off of basically um and you know maybe somebody would touch up the script before it's published I don't know but any the last thing we need to do is we just need

to add that search row Module sorry uh Google Sheets search rows module we want to search sheet two the reason why is we want to double check and basically see um whether or not we have the link from the queue in the database so sheet two what we're going to do is we're going to feed in video URL and we'll just see if this video URL is equal to one of these very first variables forever ago this one if it is equal to we're going to return it okay now assuming that we return it all

We then need to do is we just need to um update this so sheets update Arrow then all we need to do is we just need to get the same sheet YouTube content scraper update a row why is this doing it this way this is weird not really sure what happened there uh YouTube scrape content sorry I think I was referencing a folder instead of a database sheet two and then we just Need to get the row number that we just output from search row so row number and we just change the processing status to

processed now we have a Q which uh watches tasks and then we have the actual thing itself and you know we could do some other stuff like we we could technically link this specific record to that if we wanted to we we could make it as complicated as we want but I think I'll stop there this video's already gone a lot longer than I Initially thought it would uh yeah so then processed what I'm going to do is I'm just going to save this now first I'm going to connect my watch app if I runs

what I'm going to do is I'm going to um just change the data set ID so that it's the default data set ID from the first appify watch actor and then I'm going to save this and I'll copy both of these over and this will be the first blueprint which actually fires off um the Scraper so we'll we'll do this then this will be the second one which um actually runs through the whole content generator which looks freaking awesome to me I'm pretty stoked about it cool so that's that um I'm going to hopefully not

cry myself to sleep with all the tokens that I've been using obviously this sort of approaches be pretty token intensive right we're generating a boatload of content um but I hope you guys see how nuanced the content is it's not just like your typical cookie cutter AI BS especially that last section because we were using a much more advanced intelligent model to process the transcript um another thing that I think is worthwhile pointing out here is just so that we've we Chang the content like it's a little bit different from the original um you can

play around with that all you want to get is detailed and you know new as As as possible like you could I don't know you could do a ton of things you could like take the title of the video or maybe the the the summary and then feed it into perplexity and say Hey I want you to return some interesting data about this and then you could have that data come in and then you could feed it into your modification module and say Hey I want you to change the outline based off the results from

this data that might be useful for some sort of News application I'll leave it at that um but yeah had a lot of fun putting that system together if you guys any questions about the sort of system feel free to drop them down below if you have suggestions for future ones I'd love to hear them um as you guys could tell I'm getting a lot of my recent build requests directly from my audience so it means a lot to me that you guys keep on providing the ideas and I think I'll help uh sort of

Pay It Forward by Actually building them out as long as it's just something I haven't done before I want to keep this reasonably new aside from that if you could like subscribe do all that fun YouTube stuff I'll catch you on the next one really appreciate your time cheers