Claude + Facebook Ads (FULL COURSE) transcript

This is the definitive course for Claude and Facebook Ads. Across our portfolio of over 100 brands, we use Claude on a day-to-day basis to integrate directly with Facebook Ads. Everything I'm about to teach you step-by-step is from the internal processes that we've created for our clients at The Moonlighters. From my perspective, if you learn how to integrate this now, you're going to be leaps and bounds ahead of your competitors.

You'll have workflows that save you time, allow you to produce more creative fast and very cheap. You'll have anomaly detection so you don't overspend or waste your ad spend, and you'll even know when to scale your accounts. And everything is set up in a foundational way. We all know that every large language model gets better and better and better. Claude 6 months from now and a year from now is going to be 10 times stronger than it is today.

Setting the foundations now, building the skills now, building the workflows and integrations today will allow you to exponentially improve as the time goes on and as all of these systems get better. We're going to be starting at the very beginning and going step-by-step through a very advanced curriculum. First, we're going to start with the basic setup. I'm going to show you how to integrate Claude into your Facebook account right away.

This is only going to take a few minutes. Next, we're going to go into our basic capabilities. So, I'm going to show you really how to prompt this thing, how to get some good information out of it, and how to ask the questions that you might just want to ask. There's a certain way to do this and there's certain permissions that are going to help make this even easier to do. Next, we're going to get into reporting. Now, when it comes to reporting, I'm going to show you some prompts that you could use so you could build repeatable reports that allow you to see the most critical things that you're looking for throughout your business.

After that, I'm going to show you how to schedule reports. This is going to really help get a debrief on what's going on in your ad account both creatively and just from an overall business holistic perspective. From there, I'm going to go ahead and start building stuff. We're going to build anomaly detectors and winning detectors. These are going to be two things that are really going to set the foundations to let you understand how things can be built within the integration between Claude and Facebook.

Think of this as someone who's monitoring your account at regular intervals and warning you when things are going in the wrong direction. We generally want to set these up to avoid overspending or wasted ad spend in any case. Winning detection would do the exact opposite. It would tell us when we need to do more of something. Next, we're going to connect our additional business information. This is going to be connecting beyond just Facebook ads.

So, in many cases, it's going to be Shopify or whatever service provider you use. This is just getting things like cost of goods and critical information in so that Claude has more context to actually help you think through your own decisions. It's very critical here to understand that with just looking at Facebook ads in a silo, Claude is actually quite limited as to what it could understand. We have to give Claude a ton of context for it to know what to pull and know what you're actually talking about.

After this, we're going to build some fun stuff. This is where really the ball gets rolling. This is where most people will find the most value both short-term and long-term. We're going to build creative recommendation tool that's going to show you your winners and your potential winners. This is all going to be rule-based and prompt-based. There's going to be multiple prompts you need to set up. And then again, we can schedule these reports using some of the skills we learned earlier in the course.

We're also going to build creative workflows. So, this builds upon creative recommendations. And this is when we take those recommendations or potential ads and we actually build them into auto-generated AI workflows. Think of this as generating images with whatever your favorite tool is, generating scripts, generating briefs. I'm also going to talk a lot about future workflows and I think so everybody really understands this.

Future workflows are going to be the most important case for everybody. But if you don't build one through nine here, you're really going to put yourself at a massive disadvantage. In fact, you won't be able to build future workflows without building your original starting workflows. Future workflows are essentially the future of everything that I'm talking about here. This is what's going to allow you to build and build and build and really stack multiple workflows in your business that will eventually either replace employees or or allow your current team to be way more efficient.

Finally, I'm just going to cap off with current day limitations, understanding that many of these limitations are going to be completely eradicated soon because Claude's going to get better, Facebook's integration's going to get better, but right now there are some limitations that you need to be aware of because as of this moment right now, Claude is not as good as a regular media buyer or a growth director. Okay, to set up this integration, the very first thing we're going to do is go into Claude and click on this toolbox which says customize.

From here, we're going to click on connectors and then we're going to click this plus button and click add custom connector. I'm just going to call this the Facebook Ads MCP. You can call this whatever you want. And then you're going to paste in https: //mcp.facebookads.com/ads. You then click add from here and you'll see the meta or the Facebook logo automatically show up. Now, it might automatically connect for you.

If it doesn't, just click connect like so. In this case, because we use this, I just have to click reconnect. In your case, you might just need to click the actual business manager that you want to assign to Claude. And then you'll just be able to click connect. Once you do this, we now have our first look at all of our tool permissions. Now, mine are already set to custom, but just to make this really easy for you, make sure everything for now is on needs approval.

If everything's on needs approval, it means the very first time that you run something that actually needs approval, it's going to ask you if you want to move it to always allow. There are absolutely some things I'm totally comfortable always allowing and there's other things I'm totally not. For example, I don't want Claude to automatically be allowed to edit my budgets, change my ads, or change critical details. I want that always to be confirmed with me.

But I'm cool with Claude, for example, getting ad insights, which is why I have this on always allow. As a general rule of thumb, all of the insights can easily be moved to always allow, but for now, just to keep your life easy, keep it on needs approval for now, and then you're going to be able to select always allow later. So, once we've done this, we can go back to Claude and we could actually enable the MCP. So, we can click this plus button right here, hover over connectors, and you should see the Facebook Ads MCP is selected.

Now, once we do this, we just need to prompt Claude to actually see if we have access to this MCP. As of right now, around 10 to 20% of total ad accounts have access and if you don't have access, ask your Facebook rep or just request directly from support. I think this is going to be rolled out to everyone hopefully by the time this video is live. So, what you should do from this point forward now is figure out if Claude is in fact connected to the ad account.

Okay, so the very first prompt we're going to use is just to find out if we actually have access and make sure our connection was set up properly. I like to actually prompt this exact thing cuz it tends to find all the information in a really simple way. Using the Facebook Ads MCP that I just connected through the connectors, please list out all of the ad accounts that I have access to and specifically call out the ad accounts that have access to the MCP connection.

So, as we can see right away, it's starting to pull in the MCP and this is going to be an always allow. All right, so this only took a minute here and we can see I have a ridiculous 236 ad accounts connected. These are all the businesses that we work with and there's only 20 of the 236 that are not connected. So, yippee ki yay, 216 accounts are MCP enabled. So, all we're going to do for this is we're just going to pick one account and we're going to use this as our example account throughout everything.

So, for our purposes, I'm just going to be using one account. I'm going to say, let's use the account number and I find it a little bit easier to use the actual account number. It seems to be able to identify it much quicker and as you can see in this case, it's MCP enabled, queryable, we're good to go. So, now that our setup is done, let's go through some of our basic capabilities. Now, when it comes to basic capabilities, there's about a thousand and one things that you can do, but just so we understand the basis of this, we could ask anything.

And it even gives us a few starting points in this very first prompt. Now, just so we understand, things that I might ask are like, what was yesterday's ROAS and spend? And in many cases, when you do run a single prompt, you're going to get the option to either always allow or just click allow once. Just be careful in many cases. I have no problem using the ads to get ad entities as an always allow. This is just pulling information, we're not pushing anything to Facebook.

So, we can see right away for this account, we spent $960, we drove a 2.07 return on ad spend, that drove 13 purchases. I could ask anything and get results really, really quickly. A quick example of something that would be a little bit more advanced, please plot the last 30 days spend, ROAS, purchases, impressions, and frequency in a very simple table. And just like mentioned before, we're able to now see a very easy to understand, easy to comprehend table based on any of the metrics we had.

Where things start to get a little bit more advanced is when we could actually take and build upon this table. So, what I'm just going to say here is, use this data to actually build something that I can visualize and understand more easily. Make it slightly interactive, but overall keep it pretty light and simple. Now, what I've asked here is quite simple. I've said, take the data that was already plotted and visualize it.

You could do this as advanced as you would like or as simple as you would like. I just want to show you how simple it is to build something that's actually really easy for us to interact and work with. For example, here's a light interactive dashboard with the last 30 days of data. Now, you can see here, I have plots here by different charts, so super easy for me to read and understand. We can see in this case for this brand, we ran a sale, that's why ROAS gets so high, that's why spend gets so high as well.

And I can look at any critical metric here as well. So, super easy for us to grab this and pull it. Now, I'm going to pause here for a second and mention something that's really important. You could obviously play with this all day and you should. Build the custom reports that you think are most applicable to your business. But, if you want the actual prompts, the templates, the actual reporting setups that we use here at the Moonlighters, then you could always go and get them inside school.com/facebook.

That's where we share all of our internal SOPs. So, tutorials just like this, breakdowns just like this are great and really helpful. But, if you're already in a position where you're trying to up-level fast and you want to make sure you're not looking at the wrong things or missing things, then not only do we have Claude friendly stuff that is all broken down, but we also have very, very critical breakdowns that are used in more advanced reporting platforms as well.

You could steal all of that inside again school.com/facebook. There's always a link in the first comment and in the description. Now that we've built this report, we need to figure out how can we receive this report every single day. Now, there's basically two use cases for consistent reporting. First is consistent, the second is large chunks of data. So, what we're going to cover really quickly is what I would do with large chunks of data if I wanted to look back at a very large analysis of Facebook ads or really any ad platform all in one.

If I wanted to understand, you know, what went on over a long period of time, well, instead of maybe breaking things down by day, I would ask for something that is very similar, but broken down by week. I would then ask for significantly more metrics to analyze the data and I would ask Claude to start trying to find patterns in the data for me. I'll show you an example of this. I'm not going to do a 52-week window because, to be honest, it's going to take 10 minutes of me sitting here waiting.

But, if you wanted to pull something like that, you could ask something along these lines. This is great. Now, I want you to do this broken out by week for data over the last 20 weeks. After you do this, I want you to include some sort of section that actually analyzes the patterns of this data. Also, feel free to include more than just spend, rowas, purchases, impressions, and frequency. Let's go pretty advanced and detailed here.

Okay, so after just a few minutes here, we now have a fully functional lightweight reporting suite. Essentially, we could see all of our core metrics over any period of time. We could toggle all of these at any time, spend, frequency, CTR, whatever we want. We can look for anomalies throughout these. We have some of this conditionally formatted by some weeks that were really bad, some weeks that were really good. And then, it's done a pretty simple analysis.

And I think some of this analysis is pretty good. Now, I want to give everybody a good warning here. Without the context that we're going to add later on, like connecting additional business information in number seven, this doesn't have a full picture of how the business actually functions. For example, it's not taking into consideration that all of these lower weeks, this business actually thrives at a 0.8 incremental new customer return on ad spend.

Now, that's some advanced stuff right there, but it's not taking into account that even though some of these weeks look low and look really poor, it's just cuz they're focused entirely on acquiring brand new customers and not counting anything else. So, you have to take all of these analyses with a grain of salt. However, these are pretty good callouts. The business runs on promo spikes. The baseline non-promo weeks are structurally unprofitable.

That's where it's wrong. CPM has climbed steadily and materially. This is a concerning trend. Now, this is true. We've seen CPMs go from around $6, all the way up to around $10. That's a really good callout. Frequency is creeping up, so it drifted down. It's now pushed back up, and the funnel is leaking at the site, not the ad. Conversion rate, which is purchases divided by clicks, is brutal in normal weeks, frequently under 1%, and then it jumps super high during promos.

Again, it doesn't have all the the here, but these are pretty good callouts for a general run of a report. So, at this point, I think it's safe to say that we understand the basic capabilities of using Claude integrated with Facebook Ads. We also might be halfway there when it comes to basic reporting. Even though this reporting looks really, really advanced, it's actually pretty basic. You can get all this information inside of Facebook already, and there's nothing really special going on here.

It's cool to run, it feels good to run, it looks good, but if you do this every day, you're just going to be looking at the same information. There's really going to be no trend analysis going on here if we're just adding in next week, next week, next week. Where things start to get more and more valuable is when we start to schedule our reporting. So, this was the other half of reporting, which is consistent scheduled reporting.

Now, I'm going to show you how to build and stack on your reports so you can get daily digest from your ad account, which are really, really important. A lot of people build things with Claude and ChatGPT, such as daily digest that basically give them no value. They give them some weird summary of their day, what the weather's going to be like, the emails that they've gotten over the last 24 hours, but when it comes to the money that you're spending on your ads, this could be one of the most critical, if not the most critical, daily digest that you get.

Now, I encourage most people to do this inside of Slack or wherever they get their information. You could set this up anywhere. I'm just going to show you how to do the basics of it inside of Claude. So, to do this, we're going to bring ourselves back to the daily report. Excellent. Now, I just want to keep things a little bit more simple. Build a daily report just like at first. Keep the tons of metrics that you've added in there, I really like those, but just do it for this rolling month.

For example, if you're pulling this report in May, I want to see this for May to date. If you're pulling it for June, I want to see it from June 1st to the current date. This is a monthly tracker. So, what this is going to create for us essentially is what we see here, but only thinking about the current month. Okay, and now that we have our month-to-date breakdown, I'm going to give it some context on my goals. So, let me provide you with some context on my goals.

My return on ad spend target is a 1x during non-promo periods, and it's around a 4x during promo periods. All in blended, as long as the full month is tracking to a two return on ad spend, we're doing good. Let's consider this the king goal of my account, and let's include this in the reporting. Now, again, this is just an example of how we can start to interact and build this for our own business. This is technically going to be hitting some of the principles on connecting additional business information.

So, we've at this point hit up our setup, our basic capabilities, and our basic reporting. Now, it's time to build a scheduled report. So, as this prompt is completing and we're building out that report one last time, we're now going to set this up in a way where it is scheduled and sent to us every single day. So, now that we've created our first round of our reporting, and now keep in mind, do this at your will, however you want to customize your reporting, go for it.

The next thing we're just going to prompt is something simple. We just want to ask it to save this as an artifact. And what that's going to do is save it as a standalone HTML artifact that we can open, download, or share. There's going to be another build that we do from here, but just so we all understand the basis of this, this is then going to create a month-to-date ROAS tracker as we've prompted. So, whatever you ask in that report is then going to be built as a custom artifact.

An artifact in Claude's eyes is literally everything we just built, custom coded for you, and then you could edit it and adjust it at any time. We're actually not going to be using this exact artifact, and there are a hundred different ways that you can do this. do this through Claude code, you can do this through Claude co-work, and so on. I'm just going to show you the easiest way to do this for now. Understand that my expertise is not in coding.

My expertise is in Facebook ads, and this is just the easiest way to get the 99% of what we need done. I don't care about perfection. I care about getting really good data inputted so that we can make informed decisions for the advertising accounts that we manage. From this point forward, once we've created this HTML file, we're then just going to click copy right here. This is just going to copy all of this HTML base code right here.

Once we've copied our HTML, we're then going to bring this into Claude Co-work. Now, we could technically do this in Claude Code as well. Co-work is just going to be really easy. So, all we're doing is we're opening up our Claude Desktop app. If you do not have that, Google it, download it, it's free. Open it up and connect it to your Claude account. Once you do that, you can go ahead and click live artifacts here on the left-hand side.

So, this is going to be under Claude Co-work, live artifacts. Click new artifact, create with Claude. It's going to ask an auto prompt this. I want to make a live artifact, explain what they are, then figure out look at my MCP servers. It's going to see all the MCP servers that we have already set up, and it's going to probably ask me a question. So, it knows that I have the Meta Ads connection. I'm going to click one here, and we're going to do this to be a month-to-date reporting dashboard.

How will we open it? We will open this multiple times a day. It seems like it's hitting more questions. We just want to skip through this and give it the HTML. So, we're going to skip this. We're going to skip this, and we're just going to skip this. And just before this gets out of hand, I'm just going to stop Claude here, and I'm going to tell it exactly what I'm looking for. Using the Facebook Ads MCP, build this exact HTML live artifact that I've attached below.

Do this only for the account and then we're literally just pasting our entire HTML, and we're clicking go. It's then going to be able to parse this out and build this automatically for us. To save us all a few minutes here, I've already done this. And we go to our live artifacts here, we can see we now have my already built live artifact. Now, the really cool thing about this is I could do two things from this point forward.

A, I can click reload, which is going to re-pull with the live month data, or I could ask chat anything about this. So, because I've already built this, I could add or change or add layers or do whatever I want to this already. It's using all the data from the MCP, which makes it all super, super easy to modify and change. So, the next part of this naturally is getting this to be scheduled. So, we're just going to ask Claude something really simple.

This is all great. Can you please schedule this to be updated and sent to me every single day at 9:00 a.m. Let's just say for this sake, I just want this update to be in Co-work, but wherever you want it, it doesn't matter. Maybe a Slack DM, an email, it doesn't matter. We're using Eastern Time. Excellent, and we have the scheduled task. And pretty straightforward, this is completely set. So, now if we go over on the left-hand side to scheduled tasks, we'll now see that I have my month-to-date realized tracker right here.

We have all the instructions, and this will now be sent to me every single day. Now, just to show you what this looks like, I can click run now. It's then going to pull all of the data automatically. So, it's going to run the tools just like we did before, and it's going to spit back that same chart in just a few minutes. Obviously, the advantage of scheduling this is it does it. You don't even think or know that it's happening in the background.

As long as you have Claude open, the desktop app open, this is going to run. Excellent. So, we have officially wrapped up our first four sections. We know how to set this up, we understand the basic capabilities, we understand basic reporting, and we have scheduled reporting set up. These foundational 1 2 3 and 4 really lead easily into the rest of this course. Anomaly detection and winner detection are going to be one in of the same, and the important part is that we've already built the foundation for all scheduled and anomaly detection.

So, to kick off our anomaly detection, we're actually going to build this again inside of Claude Co-work. Now, you can absolutely build this in Claude Code if that's where you're more comfortable. We have these built in both for all of the brands that we work with. Just in this case, it's really easy to build this within Co-work, especially if you're not totally familiar, because it's going to act a little bit more in a user-friendly way.

So, we're going to prompt something pretty straightforward. Using the Facebook Ads MCP connector, I want you to build an anomaly detector. This should account for any significant change over two standard deviations across the brand that we are looking at. The purpose of the anomaly detector is to detect both good and bad anomalies in my ad account. I would ideally like this to be run four times per day, and specifically, we're looking at core metrics like spend, ROAS, CPA.

Oh, we also care about secondary metrics like spikes in CPMs. Okay, now this is a case where I definitely want this to be built into my Slack and DM. I want to get this notification. I want to be warned when something is actually happening that's bad. The lookback window when we should calculate the baseline, this is going to be interesting. We want to know about daily changes, but also reference 7-day, 3-day, and 14-day changes.

At what level should anomalies be detected? They should be detected probably just at the account level would be most critical, but we'll also go campaign level. What ad accounts should be monitored? And in good fashion here, it could not find the anomaly detection channel, which is because I haven't created one for this particular anomaly alert. So, I'm just going to have it DM it directly to me. Should anomalies be flagged for significance?

Yep. And then again, in good fashion, it is automatically requesting to schedule the task, so we'll go ahead and schedule that. In this case, we're running it at 6:00 a.m., 11:00 a.m., 3:00 p.m., and 7:00 p.m. Really just to call out any anomalies possible. Okay, so now that we pull into Slack, we can see this is the message I got. We have the anomaly detection, we have the account ID, the target date, and we have obviously just a sample here of all the metrics, yesterday what happened, any deltas that we've put together.

Now, I don't love this exact anomaly detection. There's way better versions of this that we can create, but I just want to give you the gist of this is what you can actually create. It is so easy at this point to just highlight this, take a screenshot, and get a different version of it that's way easier for you to actually digest. Okay, now we understand the basics of how to build any anomaly detector. This is essentially just the point where we can create any sort of pulse that's an extra set of eyes on an account to make sure things don't go haywire.

After this, we start to think about how can we make good decisions, better decisions to compound on our ad account or compound on our entire business with winner detections. So, winner detections are going to function just like anomaly detection, except they're going to go a layer down at the ad level, and they're specifically going to help us with future workflows. So, for our winner detection, what we're going to do here is we can actually use the same Facebook Ads anomaly detector, or we could just take this exact same prompt that we did last time, and we could build a new section inside of Co-work or Claude Code.

Again, I'll say this for the 100th time, use what you're most comfortable in. Whether you're using Claude Code, Claude Co-work, or even a different AI tool, the actual function of what you're doing is going to be similar, or if not exactly the same. Okay, so for this example, we're going to use the same chat that we previously were building in. The reason for this is actually quite simple. When we're using the same chat, it has all of the context of what we've already created.

Now, you don't need to do this all of the time, but in this case, we're going to be using the same ad account. We're going to be using a lot of the same skills that we've already modified. And if you have gone back and forth in this same chat to modify how you'd like things to look, what things you don't like, what metrics you'd like, then it's going to have a little bit more context on what you're building. So, for this example here, we're just going to build our second prompt.

I want you to now build a second anomaly detector. Except this time, we're going to be building it for winning ads. So, I specifically want you to look at the winning ads in my ad account, which we'll define by any ad that drives more than 5% of the total campaign spend and is above my target return on ad spend. For this example, let's say my target return on ad spend is one. As long as the ad is driving over a 1x. I also want you to rule out any sale periods which you've previously looked at before.

So, we could just ignore days completely that have massive spikes in revenue. And I'm talking like 5x over baseline. The purpose of this anomaly detector is to specifically pull out the winning ads in the ad account. So, I want a winning ads report that pulls out the ad ID and any other contextual information you can tell me about the ad. Of course, I want the baseline stats for this ad, but I'd also like for you to pull these winning ads based on 3 days of data, 7 days of data, and 14 days of data.

I'm not looking for a single day spike. I'm looking for chunks of data analysis so I can make informed decisions. Okay, now this prompt is quite long. Feel free to adjust it how you see fit. The purpose of this all in all is that we have a single prompt that's going to give us a big dump of information and then we can narrow it down from there. By the way, all of the prompts that I'm going through, way more refined, the exact artifacts, the exact prompts to build this all for you is always inside school.com/facebook.

You could join at any time. You could steal all these prompts and leave the next day. It doesn't matter. Okay, how often should this report run? Once per day. Should the report go? Yep, same as Slack. Should we define the sale period day to exclude? We'll just go here. That's fine. Okay, we can see this has finished up. Let's take a look inside of Slack real quick. So, we can see this is a massive data dump that we've gotten here.

Now, obviously this might be too much for you. It's definitely too much for me right off the bat, but it does give me a general idea of what's going on. The cool thing is I can now curate this to my specific business. Great example here. Let's say here we have our retention campaign up top here. This $50 represents 20% of the total campaign budget. So, what I would obviously just do here is I would just take this and I'd say maybe I want to break this out by campaign type.

Maybe I want to break out retention versus prospecting. So, I don't have everything grouped together here. Maybe I only want certain kinds of ads like certain kinds of formats of ads or ads that only have a specific naming convention in them. Here's the good thing that we see here. There's this persistent winners across all three windows. I think this is pretty cool. Overall, this is solid for me to understand generally what's going on.

Here's the miss that we have. We don't have links. We don't have pictures. We don't have images. So, everything you would build on top of this would be to enhance this data dive. Now, what I would actually do here is I'd actually like to see this in two different places. So, don't get me wrong. It's nice to have a report like this, but this is overwhelming, right? I would have to modify this report pretty heavily. What I'd actually like here is to build an artifact around this.

So, we're just going to tell Claude this is pretty solid. Let's now just build a persistent artifact around this. Something that I can go in and reference at any time. I think that it would also be helpful to group things by campaign. It currently looks a little bit overwhelming in its current state. Keep in mind that I'm looking for things to be sorted by amount spent. So, all we're going to now get with this is just a slightly tidied up version of what we've already built.

Okay, great example of how you can build absolutely nothing of value if you don't prompt properly. I clearly did not build the right thing. So, we're just going to ask Claude to actually just look at this for us. Okay, it looks like something has went wrong here. I've got no information here. This is completely different than the Slack report that you sent to me. So, I'd like you to take screenshots of your own artifact that you built here and try to figure out what went wrong, troubleshoot it, and create a new one.

And just like that, we now have our winning ads report. Again, modify this to your liking with the metrics that matter the most for your business. Now, I think some things that I would probably add right in here right away would be things like links to the actual ads. If we could eventually pull images, that would be huge. But really, just one of the easiest things we can do is instead of having us to then look for what is this ad, can we then take this ad and click a single link to then drive us right to the ads manager to preview it.

Right? These are just easy things that we can do that allow the entire process to be much easier for ourselves. Okay, so instead of tweaking this a hundred times over now, we're just going to keep this as is. Feel free to modify to your liking. The most important part of everything that we're about to hit is going to be what happens after. All we've set up to this point is our basic setup, our basic capabilities of Claude in the Facebook Ads MCP, how to set up and schedule reports, and how to build both anomaly detectors for negative situations and winning ads.

What we do moving forward is literally tying it all together, and this is how it becomes the agent for your business. This is how you actually grow your business fast and actually produce more for your business using Claude's code and the Facebook Ads MCP. So, now moving on to number seven, connecting additional business information. This is the context that that going to power all of the outputs. 99% of people will have already stopped watching this course at this point, so consider yourself in the 1% that's going to have a major advantage.

At this stage, our goal is to give Claude as much information and context as possible. This is going to deliver significantly stronger outputs for creative workflows and any other workflows that we build after. So, for number seven here, I'm going to load in two very important pieces. The first thing that I'm going to put in is going to be the brand guidelines, and then the second thing we're going to put in is the output of our avatar building prompt.

Brand guidelines and avatar building prompts, the exact ones that I use are always in our school community updated based on what's actually working the best right now, but you could probably put these together on your own. So, to do this, we're going to build a very basic skill inside of Claude. We're going to click a new task here, and then we're going to say, "Build a skill that you could easily reference." This is going to contain my brand guidelines and my avatars for my brand.

Then, we're going to dump all the information, so this is going to be the brand guidelines first. Then, we're going to go into another tab right here. We're going to grab this avatar building sheet, just make a quick copy of it. We usually dump our brand guidelines right here per the SOP. We drop this into a new context of Claude, and we're just going to give this in the highest model available. We're going to drop this in.

This goes through a very extensive research process. This is something that we've built for a while. It works tremendously well. Usually, this takes a few minutes to run. So, as this is working, just so everybody understands what this entire prompt does, it basically scrapes the entire internet, specifically forums and Reddit, to get critical response and drive customer avatar clusters. This is really how marketing works in 2026 and onwards, which is where you specifically talk to an avatar.

You build problem and solution-oriented creative for that avatar. And you can kind of see as this is still going here, we have stage one, which are 20 problems in the category. So, it identified the category, it found a bunch of competitors, and then it found specific problems in this category that are all over. And you could see all the different places Useful Vitamins, eBay, Goodreads, Fortune, Grooons, Team Blind, Substack, all these different places that this information and problems are pulled from.

And then we have 20 solution angles for the brand specifically. We're just using Grooons here for example. And then it's driving us different avatars that we actually have for the brand as well. So, it's putting together all the avatars for us. And again, you want to, you know, refine this to your brand for what actually fits to you. For this example, I'm just going to use this. Honestly, this gets us like 98% of the way there.

Usually, it's just some tweaking. Maybe we remove one or two avatars in total, but we usually get really, really good clusters with just the two prompts that I showed you right there. And this saves probably 2 days of work. Also, again, you could always get it inside the community. Okay, so I'm just going to copy all of this from the clustering start. We're then going to drop this into our code instance here, and then we're just going to write avatars.

I should probably understand what I'm saying here. So, this is a lot of information. I'm going to make sure we're using Opus 4.8 high here because this is going to be pretty critical. And we just want this to write the skill for us because then this skill is going to be referenced later. Again, I'll say this for the 100th time, this could be done in a hundred different ways, and there's probably a more efficient way to do it.

But, I'm just showing you the easy first step that I would take, especially if I wasn't 100% comfortable in here. So, we're just going to easily just move this through. We're going to go ahead and send it. Great. Now that this is done, this only took a few minutes here. We actually have the full skill. This is just a referenceable skill. You could do this in a few different ways, but we're just going to go ahead and save this skill.

We can now reference it at any time, or we could just tell Claude to reference the skill. Meaning, next time we ask for copy, or creative or what we should spin up next, which we're going to do in a few minutes here, we're actually going to just be able to say pull up the Gruns brand skill at any time and it will reference it. In fact, as we continue to use the same thread, which we've done a few times now, it's going to naturally reference it because it's smart enough to understand that the skill applies to the business.

So, all of that done here, we've now connected additional business information that's going to help with 8, 9, and 10. This is going to be basically the groundwork that you could build upon any time. And by the way, you could add more information, upgrade your skills at any time, and it's so easy because all you need to do is just tell Claude to upgrade, change, or modify the skill. Moving on to the next piece here, we have creative recommendations.

This is going to function very similar to our winning detection. The only difference here is we're now going to start to think about this from a workflow perspective. And we're going to be adding a lot more context than we had previously. I'm just going to ask it, "Can you pull the top spending ad for me with the ad ID and link it in Ads Manager?" Also, can you pull the actual creative so I can see it? Here's the great thing.

Whenever you ask for something like, "Can I see it?" that also should mean in most cases that Claude can see it, too. We have the Ads Manager click, so we can click this. We can then open this up right away inside of Ads Manager, which is great. We then also have the live ad preview and then the video that we can watch on Facebook. What's really cool about this is at the bottom here, it noted, "The image above is a rendered creative, a vertical before and after split showing XYZ XYZ XYZ." This is excellent.

We know that it could actually see the ad for us. Because of this, we now want to use this workflow to then start to generate more. This is where things get fun. So, here's what we're going to build, and it's very easy. Please build a basic schedule that pulls exactly this information that you have above, taking my top spending ads over the last 14 days. Do this schedule only once per week and focus on the top three highest spending ads in the account.

Ignore everything else. So, all we're building again is a basic schedule. You're going to see how this all starts to tie in together here. Now, while that's working in the background here, I want to explain the difference between winners and potential ads. It's very, very simple. You just need to define your winning parameters based on your efficiency metric and as long as that ad has spent. Typically, we like to think of winning ads as having a hit rate.

As long as both of these conditions are met, then it is in fact a winner. Potential in our case would be something like 5% of campaign budget, but missing TROAS or TCPA. Or on the flip side, hitting TROAS or TCPA, but less than 5% of the budget. Potential ads need to be tweaked. Winning ads need to be replicated. So, the workflow that we're going to now start to create is going to be based on these winning and potential parameters.

Keep in mind, I'm not narrowing down all the edge cases for every single business. Just in your case, as long as you're hitting your specific goals and targets, that's what you want to build your workflows around. Back into Claude here, we have our three ads that we pulled through. From this stage, we're now going to build on top of this one last time that's going to start auto-generating work for us. All I'm going to say here, again, in plain English, for any of these top ads that are image ads, I want you to take the image preview and build a variety of dynamic iterations of the same ad.

Feel free to change any component that you want that you think would be most important. Don't forget to reference our brand guidelines skill that we built together earlier. Every time you create these new images, I want you to review them before you send them off to me to make sure they follow the brand guidelines avatars, and feel free in a few of these variations to be unique and creative. We're not looking for tiny little tweaks or iterations, we're looking for moderately big changes based on an original ad that we already know drives a lot of spent.

Okay. So, the point of this now is as you can probably imagine, we're building workflows. So, our creative recommendations, right? This is really easy. This is just building the foundation of what's going to be recommended and what's going to be created in our workflows. Creative workflows is really where the fun begins. To put this into perspective, let's say we have all our winning ads up top here. What we want to do is every time there's a winning static ad, we want to create more iterations.

So, we have the V1, we have the V2, we have the V3, and so on. And what winds up happening is this grouping of ads becomes stronger and stronger with the winner. Because we have a positive signal based on high spend and hitting our target, and it actually expands the entire ad account, and makes the ad account more bulletproof. You'll see less fatigue in the account. It's a very positive thing that happens. In addition to our winners, we're also going to have our potential ads, which we'll call right here.

Now, for our potential ads, we're going to do the same thing. We're going to create a V1, a V2, and a V3, right? And every time we do this, every time we're creating these new versions, especially for potential ads, we're looking for variety. We're not trying to create the same ad over and over. We're trying to create different ads. What you've basically done by just doing this workflow right here is you have massively expanded your ad account.

You now have creatives that are flying in on a regular basis. You're giving these creatives 14 days to run. You're analyzing these creatives right here, and then guess what you're doing? You are creating more creatives, right? So, you're constantly building this natural creative flywheel with your winners and with your potential ads. Okay, so after you run this, you're probably going to notice you run into a big limitation.

That limitation is going to happen because you have all this context, you have all of these briefs, but Claude cannot generate images for you automatically. It could do so in regular chat. These images are often very weak. Claude is not the best image generation tool. ChatGPT tends to be a bit stronger. Nano Banana tends to be very strong as well. Now, this is the point where you get to customize your experience based on what you and your team actually like to use.

For example, my team uses Super Scale primarily. We used Higgs Field a little bit as well. And then for static images, we tend to use Nano Banana Pro. We tend to use Arc Ads as well. All of these tools can be used at your own discretion. We are a large agency. We work with a lot of brands, and because of that, we have the capability to use everything under the sun based on what's the right fit. So, at this point with the briefs that we've created here on the right side, we kind of have an idea of where things can go.

But, as I've mentioned before, we've hit a clear wall. We're not actually generating anything, and that's just because Claude itself is not very good at generating images or generating videos. There's some license to do this, but not too great. There's a whole lot of other tools that we can use that are much better. Think of just Claude as like the brain, the architect for all of this, and what we're going to use is other model context protocols, other MCPs, to tie this all together.

So, keeping this as simple as possible, I'm going to show you how we set this up in Super Scale because this is who we use. So, if you click on Super Scale down here, where the MCP is, and you literally just copy this MCP. You can then go into Claude. You can then go to your settings in Claude. You can go to connectors. Click customize right here. Click the plus button. Add a custom connector. We'll just call this Super Scale, and we'll just paste this in.

And once we just click this add button right here, we're going to have access to interact with these Super Scale MCP, aka create images, videos with context using the brain that we've already created to actually generate things for us on repeat. So, I'll just click add. So, then if we go back to our chat here, we can prompt in something very simple. Okay, this is fine. Now, since you can't generate images on your own, please use the MCP to build these images for me.

You can go back in the chat, take one of these prompts, and just build one image ad for me. I want to use Nano Banana Pro and build two variations per iteration that's necessary. Do this all in vertical formats. Now, just to show you what's going on here, we're using Higgs Field for this job. I could actually specifically tell it to use Super Scale. I could tell it to use Higgs Field. I could tell it to use anything that I've set up.

It doesn't really matter at all. As I said, I tend to get the best results from Super Scale, but this is going to just go really quick. Now, you can see these images starting to be generated on the fly here, and just like any AI-generated image, it's going to pull in in just a few minutes. Now, really I want us to understand the workflow of what we just created here because this is critical. So, first off, every single time you have a winning ad in your ad account, we are creating somewhere around two to four briefs.

These briefs are going to have context on who our avatars are, and they're going to be able to look at historic winners in the ad account, the images specifically. They're really going to be able to understand. And then we're going to pull in those briefs into either Higgs Field, Super Scale, Arc Ads, Gemini, whatever tool you like to use to generate more images. You have essentially created, as we went over already, a key creative flywheel.

Your images are now going to be able to repeat and repeat and repeat based on what's actually working, and you could run this technically anytime, although I'd recommend 7 to 14 days on a consistent cycle. Okay, so now we've created an auto-generating image machine that you could basically prompt on tap whenever you want and auto schedule it. What about scripts and briefs? Now, scripts and briefs are going to be something that you actually do more on demand, not necessarily always on a schedule, but we get also create a similar workflow for both scripts and briefs.

Now, what I like to do for all of the briefs that we follow is I like to create this as a skill as well. So, I have our UGC creator brief template here. This is for internal and external. We like to basically push this all into one spot so that we get the optimal output. And even though most of this is blurred because it's only available inside of the Facebook Ads Blueprint, just to give you like one example, we are crafting individual hooks, we're crafting individual call-outs, and we're preparing additional shot lists so that when we actually get the creative handed to us, whether it's AI creative, UGC, or in-house product studio creative, we know that we're hitting all the marks.

Briefing is critically important. If it wasn't, everybody wouldn't talk about it. So, what we like to do is we like to take this brief just like so. Then we're going to pop this back into Claude. And what we're going to run is very simple. I've just prompted here, build a skill that could easily be referenced for scripting video ads. I've then provided the full template, and then the skill was then created. Now, anytime I want to, we could edit this skill, we could reference this skill, and anytime I want to create a new script, of course, it will automatically reference it.

For example, let's say I said, "Using the Facebook Ads MCP in the account, please find my top-performing video ad, extract the transcript, and then using the skill that you've just made, write a new script and brief." So, just like all other workflows that we've created, you know what's going to happen now. It's finding the transcript, it's doing exactly what we did before, and then it's going to brief in exactly what we want in our own template that we can pass to our internal team, to our designers, to our creators, whomever we want.

This ultimately saves you hours and hours over time and time and time and time again that you have to manually create over and over briefs and briefs and briefs again. So, at this point you understand that we've created basic creative workflows, generating images, building scripts, building briefs, all of which could either be auto-generated for us depending on what tools we're using, or we can actually just use these to our advantage to brief creators, and so on.

This is generally the point of where things are good right now. But, by the time this video releases, there's going to be a hundred new things that come out as there always are. And because of that, I want you future-proof where your mindset goes on this because every week when new advancements come out, there's going to be more that we can do. This is the foundation for everything that you're going to do in the future.

Future workflows are essentially going to be limitless, but what I really want you to focus on are just a few things. So, the core workflow that we created is finding winning ads, briefing winning ads, and then of course using some tool to build more of these ads. This is all currently done pretty well with image ads, right? We'd all agree that right now AI image ads are pretty great, but when it comes to video, it's a little iffy.

So, what we like to do from this point forward in the future workflows that I think are going to be extremely valuable are how we take one single brief and multiple tools to then create video iterations and image ads all together. So, instead of just thinking of this as an image ad briefing, we would potentially take a winning ad, whether it's an image or a video, we would then brief into the tool, but we'd also brief images, and we'd also brief video, right?

And again, this is just the start of things. So, maybe we'd have one tool for images and a different tool for videos. It doesn't matter. Really, we just want to make sure on autopilot we're getting our winning ads, we're briefing in images and videos. And the cool thing is you're going to be able to modify the briefs based on what's necessary for the tool or for the concept. If it's an image concept, we know it could be simple, iterations are really easy to make.

If it's a video concept, then we might need to pre-prompt, and those are all done with skills. So, just think of skills as your intermediary between right here. So, skills are not the answer to all, but they're going to be the bridge in between the creation of the brief and what you actually get on the output. So, a skill would typically sit right here. Now, a skill in many cases is going to just be this entire workflow, but the way that we're using skills here is to interfere and be a buffer between the brief and the actual output.

What this saves for you is overspending on token cost, because spending too much on tokens is going to cost you a fortune in the long haul, especially if you're doing this on repeat. If your skills are better at filtering, you're going to save on tokens, you're going to produce better ads with higher hit rates. So, future workflows that I think are going to be great are going to be video ads, custom scripts per platform or tools, and then I think there's going to be an entire suite that we eventually layer into all of this with finance and inventory.

Now, that is TBD at the moment, there's not as many great connectors as we would like here. Shopify does port in, but the capabilities right now are a bit limited in terms of the financial role of everything. Ideally, in the long term, we create a full ecosystem that allows us to take the most basic reporting, integrate that with our creative workflow, that eventually spits back out to the finances and the inventory of the business.

On a long-term scale, we're essentially creating a massive flywheel in the business that has dozens and dozens of different components. So, while you might have your creative flywheel, for example, you could also have additional layers in here, for example, making sure that only this ad is evaluated if that ad is truly profitable based on not only the attribution of the platform, but profitable based on the unit cost or the cost of goods or the shipping threshold or the inventory available.

Right? So, you could see how we're going to eventually be able to layer on over and over and over again different things into the business to make the decision-making even better. Overall, the workflow that we've created today is the foundation and then what you build on top of this should be custom to your own business. Okay, before we wrap up, I want to hit you with some of the biggest, most critical limitations. Now, you might be thinking of limitations as something that is not available today and it's going to be available tomorrow.

That is one way to think about it, but think of this more as a red flag. Some people are going to run with Claude in the Facebook Ads integration, upload all their ads tomorrow, they're going to create massive flywheels, and their ad accounts are going to get bloated really quickly. They're going to have a bunch of crappy ads that don't perform or they might have a few good ads, but they never get adequate spend. And this is going to be a death loop for most businesses.

Tools like this are fantastic at enhancing what you're currently doing. They will make you more efficient if done right, and if you build your workflows custom to yourself, they will really enhance your entire organization. So, what we do here at The Moon Liters is we use everything that I just showed you as a filter, almost like a step one into just the foundation of the critical mass or media buying is still very, very relevant.

The most important thing to understand is that at the end of the day when it comes to running Facebook Ads, you're spending thousands and thousands of dollars, likely of your very own money, you need to make sure you have a human filter at the end of that. Just because everything looks sexy and shiny right now does not mean that it's the appropriate thing to do for the business today. So, be careful, limit yourself, don't overdo it in your ad account, and most importantly, do not remove the most important filter from making sure your cash is used properly.

That is everything that I'm going through today in the Claude Meta Ads MCP tutorial. This is the start of a new era, and if you want to see everything I just went through, not only in this format, but the exact prompts, all of the briefs I showed you, everything laid out in a very easy step-by-step version, then we've added all of this to school.com/facebook. You can join there anytime, or if you want all this handled for you, if you don't want to ever have to deal with this, you can go to the moonlighters.com/apply to see if you're a fit to work with us.

That's all for today. I hope you got a ton of value out of this, and I'll see you in the next one.