Claude Code can now create 100s of winning meta ads (in 9 mins)

In this video, I'm going to show you how you can use Claude Code to scrape your competitors' winning ads and then replicate them with your own branding and your own product. And AI can even tell you which ones are most likely to perform better. Just quickly, if you don't know who I am, my name is Rory Ridgers and I run an agency called Autonomously AI where we find and eliminate operational bottlenecks using AI. So if that's something you might be interested in, I'll leave a link in the description of this video and you can book up a call with me. Without further ado, let's jump into the video. So from a high level, the way this system works is it simply scrapes the Meta Ads library and then analyzes the top performing ads of all of your competitors and then simply replicates that using your own brand guidelines. So your own colors, your own logo, your own product. First of all, I want to show you inside of this Claude Code workspace that we're going to be doing all of this out. If we come over in here, we go to the.claud files, we can find the commands and the skills. And now the skills, the ad analyzer skill is the first skill. And what this does is it uses a vision model that is able to analyze images and it's able to analyze all the images that your competitors are using in their ads. And then the Ad Library Scraper does exactly what it says on the tin. It's able to scrape the Meta Ads library for pretty much anything you want. But of course, we're gonna be using it to scrape our competitors. And then the GPT Image Generator, we're gonna be using GPT V2 image model for this. So that skill just simply allows Claude Code to be able to better use that image model. Then we have a few different commands that we're gonna run through. The first one is analyze. That is going to analyze all of your competitors on Meta Ads. A lot of them incorporate these skills as well, so we don't run the skills separately. We run the commands., and then Claude will use the skills as it sees fit. So then over in the skills, the first one that we're gonna be looking at is scrape.md, and this allows you to scrape meta ads from your competitors. And by the way, the way the skills and the commands work, a lot of the commands prompt Claude Code to use the skills. So we don't use the skills separately. We run the commands and then the skills are used during the commands. Then after scrape, we have analyze, so we can actually analyze what the top performing ones are out of all of those ads after we scrape them. And then we can generate ads, so actually generate our own ads based on the top performing ones from our competitors. And then we have review, which basically generates that landing page that I just showed you, which makes it look all nice and is easy for us to view. Then another thing I want to show you is this config file where the main two things we have in it are brand and then competitors. These are just TypeScript files. So first of all, brand, we have our brand, which we're going to be doing a supplement brand specifically selling, uh, protein powder. So it's going to be called Apex Fuel. Our product name is Apex Whey Isolate. And then this is the one line about what we do. And then we just have like our colors in here. Our fonts, our audiences, CTAs, and also our voice, and then also some banned terms that we don't want to use. And then also we have competitors in here. So these competitors are just selling the same stuff. We've already got their Meta Ads page ID, just makes things run a little bit more smoothly. But of course you can just get Claude Code to grab them for you, you don't have to find them yourself. And then the only other thing I want to show you is inside of Apify here. Apify is what we're going to be using to actually scrape the Meta Ads library. Specifically, we're going to be using this actor here, so the Facebook Ads Library Scraper. Cost $5 per 1,000 ads, so it's pretty cheap because based on those 1,000 ads, you could create 10,000 of your own ads if you like. This is just basically the data that we're going to use to be able to create our own ads, so you don't have to scrape a crazy amount. And the way you actually connect this up to Claude Code is if you literally just give it the link from Apify here, say I want to connect up this actor, and then give it your Apify API key. Claude Code will just handle the integration for you. So you can see if we go to this EMV example page, the only things we need to put in here is our OpenAI API key for the image generation and then our Apify API token to be to use that scraper there. So we're just gonna jump right in and actually test that out. The first command that we're gonna run here is going to be /scrape. So we're gonna run that. It already has our competitors preloaded and it already has our brand preloaded. So it's gonna know, you can see it uses a scrape skill there to be able to know how to actually scrape stuff. So prompt it to basically use the Apify Actor that I just showed you, and then it should load our competitors and our brand. So it's gonna know basically who we are and who our competitors are. So it knows what it's actually meant to be scraping. So just from testing this out, it said that basically like cache run. So the last time we ran this, that data's already still in here. So it just asks if we want to use that. But for the sake of this video, we're gonna rerun it. So I just said no rerun Apify, and then we're going to basically be able to see it inside of Apify, these actors running. So when that occurs, I will show you. So inside of Apify here, you can see it's already done about 3 runs of this actor. It's just doing the 4th one here, so it should almost be done. But each of them are getting about 20 to 30 results, so it's gonna do it all the way up to 100. Okay, cool. So it did 99. And like I said, this sample size doesn't really matter too much. Obviously the more data you you put in, the better the results of your ads are going to be. But from 99, if we wanted to, we could create thousands of ads. So don't think just because we've only scraped 99 that we are limited to only creating 99 ads. But for the purpose of this video, I thought it'd be easier to just scrape around 100. Then that costs us $0.57. So if you want to do like 1,000 or 990, I guess it would be around $5 to $6. Cool. So now Claude Code has basically scraped 99 ads from our competitors on Meta ads. So now what we're going to run is analyze. And what this is going to do is basically take all of those ads, going to use the vision model to look at all of the images. And then what it's going to do is that one of the best indicators to see if ads are performing well is simply just how long they've been running. So if there was an ad that was running for 5 days and then got cut and they didn't start running it again, chances are that ad probably didn't perform very well. But if there's been an ad that's been running for 2 months straight and they haven't changed it, chances are that ad has been performing really well, because no company would keep an unprofitable ad running for 2 months. That's kind of like the main indicator that we used to see what is actually doing well and what isn't. And then from there, Claude Coke basically build up a pattern of the ads that have been running longer, which means they are likely profitable and better than the other ones. And it'll be able to recognize the patterns in that, and then of course help us when we're generating our own ones. Okay, cool. So the analysis has been done, and this is basically what it's given us. So first of all is the archetype, which is basically format of the ads. We've got hero product, product on color, UGC mockup, split screen, and then text only. And it's saying the hero product is the best performing out of all of them. So it scores them, says the frequency, so how often this appears, how many brands are using it out of all the ones that we scraped, the average days that one runs, and then the maximum amount of days it runs, and then the amount of variations it sees. So you can see out of all of these, this one is clearly an outlier. It's used way more often and it's run for way longer than the other ones on average. So from that, it's basically deduced that hero product is the best, product on color second best, UGC mockups the third best, split screen is the fourth best, and then text only is the fifth best. So now from here, what we do— by the way, as well, that costs 13 cents to do 99 ad sets. So 13 cents to do 99. If you're going to do 1,000, you know, it costs you $1.30. So really not too expensive to run at all. So now all we have to do is simply do generate, and what it's going to do, it's going to take those ads, so it's best performing ads, it's going to do 2 from each of the top 2. It's just going to give us 4 ads. That's just for the sake of this video. You could get to generate as many as you want. Obviously, as you can imagine, just the more you generate, the slightly more expensive it's going to be and also the slightly longer it's going to take. But yeah, as you can see, it's going to generate 2 of the top 2 variations. So 2 of these and then 2 of these, because it knows that these 2 are the best out of all of the different ad formats that our competitors are running. So it's going to generate images for us. It's going to give us a score on how likely it thinks this ad is going to do well compared to the other ones. And it's also going to give us like the title and the description, everything like that. So it's easy to just copy and paste, put straight into your Meta ads. Okay, cool. So it's generated all of those ads for us. It cost us 8 cents. So I mean, altogether, including the scrapes, the scrapes is kind of like a flat cost that you only have to do one time. You can generate as many ads as you want to from it. But I mean, all in all, this has cost us less than a dollar to generate these. And if it was going to generate 1,000, it would still be, you know, under $10 to do all of that. So now that it has all of these images, it's got the images here and it's also got the captions and the titles and also the actual analysis of each of them, we're just going to run /review, which is going to generate it in that nice pretty landing page that I showed you at the beginning of the video. And that way it's going to be easy for us to visualize and also see the analysis. Okay, so here we go. This is what it generated. So these first two here are hero product, and then these ones are product on color. So if we open this up, this one's an 88% performance confidence, so it's 88% sure it's going to perform pretty well or better than the other ones., and then here is why it's confident it does it. It's basically a bigger version of here. I'll just read out this part, which is it was used by 4 of the scraped brands, 56 ads in a cluster, max active lifespan is about 154 days, 21 distinct variant groups, and then lead brand was Transparent Labs, and then average per ad score is 85 out of 100 out of all the ones that it scraped. Then it just gives us a headline and then a subheadline here as well. We can just copy this. Take it straight over to Meta Ads. And this one is going to be pretty similar, but it's kind of just like a variation of this image. And then this one, a little bit different. This one's a 55 out of 100, and this is mainly just because it was used less from all of our competitors. So the maximum lifespan was about the same, but then the average lifespan I think was a little bit lower from what I remember. So that is why obviously the score is lower. So then yeah, like I said, you could generate hundreds if you wanted to. I just did 4 for the example. In this video. All of these are obviously very professional looking ads. I mean, that looks like an ad, you know, the product looks real, looks like a proper photo shoot, and it looks like it's actually been made by a graphic designer. And it's actually backed up by data on what our competitors are doing and what we know is working well for them. All right, that is it for the video, guys. I hope you enjoyed it. As I mentioned at the beginning of the video, if you were looking to implement AI systems like this into your business, then follow the link in the description of this video. You can book up a call with me and we can have a chat about it. But if not, that's it for the video. See you in the next one.