YT YouTube2Text

YouTube2Text → Transcripts → Claude + Facebook Ads (FULL COURSE)

Claude + Facebook Ads (FULL COURSE) — Transcript

by Sam Piliero · 10,772 words · 1,536 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00This is the definitive course for Claude
0:02and Facebook Ads. Across our portfolio
0:04of over 100 brands, we use Claude on a
0:07day-to-day basis to integrate directly
0:09with Facebook Ads. Everything I'm about
0:11to teach you step-by-step is from the
0:12internal processes that we've created
0:14for our clients at The Moonlighters.
0:16From my perspective, if you learn how to
0:18integrate this now, you're going to be
0:20leaps and bounds ahead of your
0:21competitors. You'll have workflows that
0:22save you time, allow you to produce more
0:24creative fast and very cheap. You'll
0:26have anomaly detection so you don't
0:28overspend or waste your ad spend, and
0:30you'll even know when to scale your
0:32accounts. And everything is set up in a
0:33foundational way. We all know that every
0:36large language model gets better and
0:37better and better. Claude 6 months from
0:39now and a year from now is going to be
0:4110 times stronger than it is today.
0:44Setting the foundations now, building
0:46the skills now, building the workflows
0:48and integrations today will allow you to
0:51exponentially improve as the time goes
0:54on and as all of these systems get
0:56better. We're going to be starting at
0:58the very beginning and going
0:59step-by-step through a very advanced
1:02curriculum. First, we're going to start
1:03with the basic setup. I'm going to show
1:05you how to integrate Claude into your
1:07Facebook account right away. This is
1:08only going to take a few minutes. Next,
1:10we're going to go into our basic
1:12capabilities. So, I'm going to show you
1:13really how to prompt this thing, how to
1:15get some good information out of it, and
1:17how to ask the questions that you might
1:18just want to ask. There's a certain way
1:20to do this and there's certain
1:21permissions that are going to help make
1:23this even easier to do. Next, we're
1:24going to get into reporting. Now, when
1:26it comes to reporting, I'm going to show
1:27you some prompts that you could use so
1:29you could build repeatable reports that
1:31allow you to see the most critical
1:33things that you're looking for
1:34throughout your business. After that,
1:36I'm going to show you how to schedule
1:37reports. This is going to really help
1:39get a debrief on what's going on in your
1:40ad account both creatively and just from
1:43an overall business holistic
1:44perspective. From there, I'm going to go
1:45ahead and start building stuff. We're
1:47going to build anomaly detectors and
1:49winning detectors. These are going to be
1:51two things that are really going to set
1:52the foundations to let you understand
1:54how things can be built within the
1:57integration between Claude and Facebook.
1:59Think of this as someone who's
2:00monitoring your account at regular
2:02intervals and warning you when things
2:04are going in the wrong direction. We
2:05generally want to set these up to avoid
2:07overspending or wasted ad spend in any
2:10case. Winning detection would do the
2:11exact opposite. It would tell us when we
2:13need to do more of something. Next,
2:15we're going to connect our additional
2:16business information. This is going to
2:18be connecting beyond just Facebook ads.
2:21So, in many cases, it's going to be
2:22Shopify or whatever service provider you
2:25use. This is just getting things like
2:26cost of goods and critical information
2:28in so that Claude has more context to
2:31actually help you think through your own
2:33decisions. It's very critical here to
2:35understand that with just looking at
2:36Facebook ads in a silo, Claude is
2:38actually quite limited as to what it
2:40could understand. We have to give Claude
2:42a ton of context for it to know what to
2:44pull and know what you're actually
2:45talking about. After this, we're going
2:47to build some fun stuff. This is where
2:48really the ball gets rolling. This is
2:50where most people will find the most
2:52value both short-term and long-term.
2:54We're going to build creative
2:55recommendation tool that's going to show
2:56you your winners and your potential
2:58winners. This is all going to be
2:59rule-based and prompt-based. There's
3:01going to be multiple prompts you need to
3:02set up. And then again, we can schedule
3:04these reports using some of the skills
3:05we learned earlier in the course. We're
3:07also going to build creative workflows.
3:08So, this builds upon creative
3:10recommendations. And this is when we
3:11take those recommendations or potential
3:13ads and we actually build them into
3:15auto-generated AI workflows. Think of
3:18this as generating images with whatever
3:20your favorite tool is, generating
3:22scripts, generating briefs. I'm also
3:23going to talk a lot about future
3:25workflows and I think so everybody
3:26really understands this. Future
3:28workflows are going to be the most
3:29important case for everybody. But if you
3:31don't build one through nine here,
3:33you're really going to put yourself at a
3:34massive disadvantage. In fact, you won't
3:37be able to build future workflows
3:38without building your original starting
3:41workflows. Future workflows are
3:42essentially the future of everything
3:44that I'm talking about here. This is
3:46what's going to allow you to build and
3:47build and build and really stack
3:50multiple workflows in your business that
3:52will eventually either replace employees
3:54or or allow your current team to be way
3:56more efficient. Finally, I'm just going
3:58to cap off with current day limitations,
4:00understanding that many of these
4:01limitations are going to be completely
4:03eradicated soon because Claude's going
4:05to get better, Facebook's integration's
4:07going to get better, but right now there
4:09are some limitations that you need to be
4:10aware of because as of this moment right
4:12now, Claude is not as good as a regular
4:15media buyer or a growth director. Okay,
4:17to set up this integration, the very
4:19first thing we're going to do is go into
4:21Claude and click on this toolbox which
4:23says customize. From here, we're going
4:25to click on connectors and then we're
4:27going to click this plus button and
4:28click add custom connector. I'm just
4:30going to call this the Facebook Ads MCP.
4:32You can call this whatever you want. And
4:33then you're going to paste in https:
4:35//mcp.facebookads.com/ads.
4:38You then click add from here
4:42and you'll see the meta or the Facebook
4:45logo automatically show up. Now, it
4:47might automatically connect for you. If
4:49it doesn't, just click connect like so.
4:50In this case, because we use this, I
4:52just have to click reconnect. In your
4:54case, you might just need to click the
4:56actual business manager that you want to
4:58assign to Claude. And then you'll just
4:59be able to click connect. Once you do
5:01this, we now have our first look at all
5:03of our tool permissions. Now, mine are
5:05already set to custom, but just to make
5:07this really easy for you, make sure
5:08everything for now is on needs approval.
5:11If everything's on needs approval, it
5:13means the very first time that you run
5:15something that actually needs approval,
5:17it's going to ask you if you want to
5:19move it to always allow. There are
5:21absolutely some things I'm totally
5:22comfortable always allowing and there's
5:24other things I'm totally not. For
5:26example, I don't want Claude to
5:27automatically be allowed to edit my
5:29budgets, change my ads, or change
5:31critical details. I want that always to
5:33be confirmed with me. But I'm cool with
5:35Claude, for example, getting ad
5:37insights, which is why I have this on
5:38always allow. As a general rule of
5:40thumb, all of the insights can easily be
5:42moved to always allow, but for now, just
5:44to keep your life easy, keep it on needs
5:46approval for now, and then you're going
5:48to be able to select always allow later.
5:50So, once we've done this, we can go back
5:52to Claude and we could actually enable
5:54the MCP. So, we can click this plus
5:56button right here, hover over
5:57connectors, and you should see the
5:59Facebook Ads MCP is selected. Now, once
6:02we do this, we just need to prompt
6:03Claude to actually see if we have access
6:07to this MCP. As of right now, around 10
6:09to 20% of total ad accounts have access
6:12and if you don't have access, ask your
6:14Facebook rep or just request directly
6:16from support. I think this is going to
6:18be rolled out to everyone hopefully by
6:19the time this video is live. So, what
6:21you should do from this point forward
6:23now is figure out if Claude is in fact
6:26connected to the ad account. Okay, so
6:28the very first prompt we're going to use
6:30is just to find out if we actually have
6:32access and make sure our connection was
6:34set up properly. I like to actually
6:36prompt this exact thing cuz it tends to
6:38find all the information in a really
6:40simple way. Using the Facebook Ads MCP
6:42that I just connected through the
6:44connectors, please list out all of the
6:47ad accounts that I have access to and
6:49specifically call out the ad accounts
6:51that have access to the MCP connection.
6:56So, as we can see right away, it's
6:57starting to pull in the MCP and this is
6:59going to be an always allow. All right,
7:01so this only took a minute here and we
7:03can see I have a ridiculous 236 ad
7:06accounts connected. These are all the
7:07businesses that we work with and there's
7:09only 20 of the 236 that are not
7:12connected. So, yippee ki yay, 216
7:15accounts are MCP enabled. So, all we're
7:17going to do for this is we're just going
7:19to pick one account and we're going to
7:20use this as our example account
7:22throughout everything. So, for our
7:23purposes, I'm just going to be using one
7:25account. I'm going to say, let's use the
7:26account number and I find it a little
7:28bit easier to use the actual account
7:30number. It seems to be able to identify
7:32it much quicker and as you can see in
7:34this case, it's MCP enabled, queryable,
7:36we're good to go. So, now that our setup
7:38is done, let's go through some of our
7:39basic capabilities. Now, when it comes
7:41to basic capabilities, there's about a
7:43thousand and one things that you can do,
7:45but just so we understand the basis of
7:48this, we could ask anything. And it even
7:50gives us a few starting points in this
7:52very first prompt. Now, just so we
7:54understand, things that I might ask are
7:55like, what was yesterday's ROAS and
7:59spend?
8:02And in many cases, when you do run a
8:04single prompt, you're going to get the
8:05option to either always allow or just
8:07click allow once. Just be careful in
8:10many cases. I have no problem using the
8:12ads to get ad entities as an always
8:15allow. This is just pulling information,
8:17we're not pushing anything to Facebook.
8:19So, we can see right away for this
8:20account, we spent $960, we drove a 2.07
8:23return on ad spend, that drove 13
8:25purchases. I could ask anything and get
8:27results really, really quickly. A quick
8:30example of something that would be a
8:31little bit more advanced, please plot
8:33the last 30 days spend, ROAS, purchases,
8:37impressions, and frequency in a very
8:39simple table.
8:42And just like mentioned before, we're
8:44able to now see a very easy to
8:45understand, easy to comprehend table
8:47based on any of the metrics we had.
8:49Where things start to get a little bit
8:50more advanced is when we could actually
8:51take and build upon this table. So, what
8:54I'm just going to say here is, use this
8:56data to actually build something that I
8:58can visualize and understand more
9:00easily. Make it slightly interactive,
9:02but overall keep it pretty light and
9:03simple.
9:04Now, what I've asked here is quite
9:06simple. I've said, take the data that
9:07was already plotted and visualize it.
9:10You could do this as advanced as you
9:11would like or as simple as you would
9:13like. I just want to show you how simple
9:14it is to build something that's actually
9:15really easy for us to interact and work
9:18with. For example, here's a light
9:20interactive dashboard with the last 30
9:22days of data. Now, you can see here, I
9:24have plots here by different charts, so
9:26super easy for me to read and
9:27understand. We can see in this case for
9:29this brand, we ran a sale, that's why
9:30ROAS gets so high, that's why spend gets
9:32so high as well. And I can look at any
9:34critical metric here as well. So, super
9:36easy for us to grab this and pull it.
9:39Now, I'm going to pause here for a
9:40second and mention something that's
9:41really important. You could obviously
9:43play with this all day and you should.
9:45Build the custom reports that you think
9:46are most applicable to your business.
9:49But, if you want the actual prompts, the
9:51templates, the actual reporting setups
9:53that we use here at the Moonlighters,
9:55then you could always go and get them
9:56inside school.com/facebook.
9:59That's where we share all of our
10:00internal SOPs. So, tutorials just like
10:03this, breakdowns just like this are
10:05great and really helpful. But, if you're
10:07already in a position where you're
10:08trying to up-level fast and you want to
10:10make sure you're not looking at the
10:11wrong things or missing things, then not
10:13only do we have Claude friendly stuff
10:15that is all broken down, but we also
10:17have very, very critical breakdowns that
10:20are used in more advanced reporting
10:22platforms as well. You could steal all
10:24of that inside again
10:25school.com/facebook. There's always a
10:28link in the first comment and in the
10:29description. Now that we've built this
10:31report, we need to figure out how can we
10:33receive this report every single day.
10:36Now, there's basically two use cases for
10:37consistent reporting. First is
10:39consistent, the second is large chunks
10:42of data. So, what we're going to cover
10:43really quickly is what I would do with
10:45large chunks of data if I wanted to look
10:47back at a very large analysis of
10:50Facebook ads or really any ad platform
10:52all in one. If I wanted to understand,
10:54you know, what went on over a long
10:55period of time, well, instead of maybe
10:57breaking things down by day, I would ask
10:59for something that is very similar, but
11:01broken down by week. I would then ask
11:03for significantly more metrics to
11:05analyze the data and I would ask Claude
11:07to start trying to find patterns in the
11:10data for me. I'll show you an example of
11:12this. I'm not going to do a 52-week
11:14window because, to be honest, it's going
11:15to take 10 minutes of me sitting here
11:17waiting. But, if you wanted to pull
11:18something like that, you could ask
11:20something along these lines. This is
11:21great. Now, I want you to do this broken
11:24out by week for data over the last 20
11:27weeks. After you do this, I want you to
11:29include some sort of section that
11:30actually analyzes the patterns of this
11:32data. Also, feel free to include more
11:35than just spend, rowas, purchases,
11:36impressions, and frequency. Let's go
11:38pretty advanced and detailed here.
11:44Okay, so after just a few minutes here,
11:47we now have a fully functional
11:49lightweight reporting suite.
11:50Essentially, we could see all of our
11:52core metrics over any period of time. We
11:55could toggle all of these at any time,
11:58spend, frequency, CTR, whatever we want.
12:00We can look for anomalies throughout
12:02these. We have some of this
12:03conditionally formatted by some weeks
12:05that were really bad, some weeks that
12:07were really good. And then, it's done a
12:09pretty simple analysis. And I think some
12:12of this analysis is pretty good. Now, I
12:14want to give everybody a good warning
12:15here. Without the context that we're
12:17going to add later on, like connecting
12:19additional business information in
12:21number seven, this doesn't have a full
12:23picture of how the business actually
12:25functions. For example, it's not taking
12:26into consideration that all of these
12:28lower weeks, this business actually
12:30thrives at a 0.8 incremental new
12:33customer return on ad spend. Now, that's
12:34some advanced stuff right there, but
12:36it's not taking into account that even
12:37though some of these weeks look low and
12:40look really poor, it's just cuz they're
12:42focused entirely on acquiring brand new
12:44customers and not counting anything
12:46else. So, you have to take all of these
12:48analyses with a grain of salt. However,
12:51these are pretty good callouts. The
12:53business runs on promo spikes. The
12:55baseline non-promo weeks are
12:57structurally unprofitable. That's where
12:59it's wrong. CPM has climbed steadily and
13:01materially. This is a concerning trend.
13:03Now, this is true. We've seen CPMs go
13:05from around $6, all the way up to around
13:08$10. That's a really good callout.
13:10Frequency is creeping up, so it drifted
13:12down. It's now pushed back up, and the
13:14funnel is leaking at the site, not the
13:16ad. Conversion rate, which is purchases
13:18divided by clicks, is brutal in normal
13:19weeks, frequently under 1%, and then it
13:22jumps super high during promos. Again,
13:24it doesn't have all the the here, but
13:25these are pretty good callouts for a
13:27general run of a report. So, at this
13:29point, I think it's safe to say that we
13:30understand the basic capabilities of
13:32using Claude integrated with Facebook
13:34Ads. We also might be halfway there when
13:36it comes to basic reporting. Even though
13:39this reporting looks really, really
13:41advanced, it's actually pretty basic.
13:43You can get all this information inside
13:44of Facebook already, and there's nothing
13:46really special going on here. It's cool
13:48to run, it feels good to run, it looks
13:51good, but if you do this every day,
13:52you're just going to be looking at the
13:53same information. There's really going
13:55to be no trend analysis going on here if
13:58we're just adding in next week, next
14:00week, next week. Where things start to
14:01get more and more valuable is when we
14:03start to schedule our reporting. So,
14:04this was the other half of reporting,
14:06which is consistent scheduled reporting.
14:08Now, I'm going to show you how to build
14:10and stack on your reports so you can get
14:12daily digest from your ad account, which
14:14are really, really important. A lot of
14:15people build things with Claude and
14:17ChatGPT, such as daily digest that
14:19basically give them no value. They give
14:21them some weird summary of their day,
14:23what the weather's going to be like, the
14:25emails that they've gotten over the last
14:2724 hours, but when it comes to the money
14:29that you're spending on your ads, this
14:30could be one of the most critical, if
14:33not the most critical, daily digest that
14:36you get. Now, I encourage most people to
14:38do this inside of Slack or wherever they
14:40get their information. You could set
14:42this up anywhere. I'm just going to show
14:44you how to do the basics of it inside of
14:46Claude. So, to do this, we're going to
14:47bring ourselves back to the daily
14:49report. Excellent. Now, I just want to
14:51keep things a little bit more simple.
14:52Build a daily report just like at first.
14:55Keep the tons of metrics that you've
14:58added in there, I really like those, but
14:59just do it for this rolling month. For
15:02example, if you're pulling this report
15:03in May, I want to see this for May to
15:06date. If you're pulling it for June, I
15:08want to see it from June 1st to the
15:10current date. This is a monthly tracker.
15:13So, what this is going to create for us
15:14essentially is what we see here, but
15:16only thinking about the current month.
15:18Okay, and now that we have our
15:19month-to-date breakdown, I'm going to
15:21give it some context on my goals. So,
15:23let me provide you with some context on
15:25my goals. My return on ad spend target
15:27is a 1x during non-promo periods, and
15:31it's around a 4x during promo periods.
15:34All in blended, as long as the full
15:36month is tracking to a two return on ad
15:38spend, we're doing good. Let's consider
15:39this the king goal of my account, and
15:42let's include this in the reporting.
15:44Now, again, this is just an example of
15:46how we can start to interact and build
15:49this for our own business. This is
15:51technically going to be hitting some of
15:52the principles on connecting additional
15:54business information. So, we've at this
15:56point hit up our setup, our basic
15:57capabilities, and our basic reporting.
16:00Now, it's time to build a scheduled
16:02report.
16:10So, as this prompt is completing and
16:12we're building out that report one last
16:14time, we're now going to set this up in
16:15a way where it is scheduled and sent to
16:18us every single day. So, now that we've
16:20created our first round of our
16:22reporting, and now keep in mind, do this
16:24at your will, however you want to
16:25customize your reporting, go for it. The
16:28next thing we're just going to prompt is
16:30something simple. We just want to ask it
16:31to save this as an artifact. And what
16:34that's going to do is save it as a
16:35standalone HTML artifact that we can
16:38open, download, or share. There's going
16:40to be another build that we do from
16:42here, but just so we all understand the
16:44basis of this, this is then going to
16:46create a month-to-date ROAS tracker as
16:50we've prompted. So, whatever you ask in
16:52that report is then going to be built as
16:54a custom artifact. An artifact in
16:56Claude's eyes is literally everything we
16:58just built, custom coded for you, and
17:00then you could edit it and adjust it at
17:03any time. We're actually not going to be
17:04using this exact artifact, and there are
17:06a hundred different ways that you can do
17:08this. do this through Claude code, you
17:10can do this through Claude co-work, and
17:12so on. I'm just going to show you the
17:13easiest way to do this for now.
17:15Understand that my expertise is not in
17:17coding. My expertise is in Facebook ads,
17:19and this is just the easiest way to get
17:22the 99%
17:23of what we need done. I don't care about
17:26perfection. I care about getting really
17:28good data inputted so that we can make
17:31informed decisions for the advertising
17:33accounts that we manage. From this point
17:35forward, once we've created this HTML
17:37file, we're then just going to click
17:39copy right here. This is just going to
17:40copy all of this HTML base code right
17:43here. Once we've copied our HTML, we're
17:45then going to bring this into Claude
17:47Co-work. Now, we could technically do
17:49this in Claude Code as well. Co-work is
17:51just going to be really easy. So, all
17:54we're doing is we're opening up our
17:55Claude Desktop app. If you do not have
17:58that, Google it, download it, it's free.
18:01Open it up and connect it to your Claude
18:02account. Once you do that, you can go
18:04ahead and click live artifacts here on
18:06the left-hand side. So, this is going to
18:07be under Claude Co-work, live artifacts.
18:10Click new artifact, create with Claude.
18:12It's going to ask an auto prompt this. I
18:14want to make a live artifact, explain
18:16what they are, then figure out look at
18:18my MCP servers. It's going to see all
18:20the MCP servers that we have already set
18:22up, and it's going to probably ask me a
18:23question. So, it knows that I have the
18:25Meta Ads connection. I'm going to click
18:26one here, and we're going to do this to
18:29be a month-to-date reporting dashboard.
18:33How will we open it? We will open this
18:35multiple times a day. It seems like it's
18:37hitting more questions. We just want to
18:38skip through this and give it the HTML.
18:40So, we're going to skip this. We're
18:41going to skip this, and we're just going
18:43to skip this. And just before this gets
18:46out of hand, I'm just going to stop
18:48Claude here, and I'm going to tell it
18:50exactly what I'm looking for. Using the
18:52Facebook Ads MCP, build this exact HTML
18:56live artifact that I've attached below.
18:58Do this only for the account
19:01and then we're literally just pasting
19:03our entire HTML, and we're clicking go.
19:06It's then going to be able to parse this
19:07out and build this automatically for us.
19:10To save us all a few minutes here, I've
19:12already done this. And we go to our live
19:14artifacts here, we can see we now have
19:17my already built live artifact. Now, the
19:20really cool thing about this is I could
19:22do two things from this point forward.
19:24A, I can click reload, which is going to
19:26re-pull with the live month data, or I
19:29could ask chat anything about this. So,
19:32because I've already built this, I could
19:33add or change or add layers or do
19:36whatever I want to this already. It's
19:39using all the data from the MCP, which
19:41makes it all super, super easy to modify
19:44and change. So, the next part of this
19:46naturally is getting this to be
19:48scheduled. So, we're just going to ask
19:49Claude something really simple. This is
19:50all great. Can you please schedule this
19:53to be updated and sent to me every
19:56single day at 9:00 a.m.
19:59Let's just say for this sake, I just
20:01want this update to be in Co-work, but
20:03wherever you want it, it doesn't matter.
20:04Maybe a Slack DM, an email, it doesn't
20:06matter. We're using Eastern Time.
20:07Excellent, and we have the scheduled
20:09task. And pretty straightforward, this
20:10is completely set. So, now if we go over
20:13on the left-hand side to scheduled
20:15tasks, we'll now see that I have my
20:17month-to-date realized tracker right
20:19here. We have all the instructions, and
20:21this will now be sent to me every single
20:23day. Now, just to show you what this
20:24looks like, I can click run now. It's
20:26then going to pull all of the data
20:27automatically. So, it's going to run the
20:28tools just like we did before, and it's
20:31going to spit back that same chart in
20:33just a few minutes. Obviously, the
20:34advantage of scheduling this is it does
20:36it. You don't even think or know that
20:38it's happening in the background. As
20:39long as you have Claude open, the
20:41desktop app open, this is going to run.
20:44Excellent. So, we have officially
20:46wrapped up our first four sections. We
20:48know how to set this up, we understand
20:50the basic capabilities, we understand
20:52basic reporting, and we have scheduled
20:53reporting set up. These foundational 1 2
20:563 and 4 really lead easily into the rest
21:00of this course. Anomaly detection and
21:02winner detection are going to be one in
21:04of the same, and the important part is
21:06that we've already built the foundation
21:08for all scheduled and anomaly detection.
21:12So, to kick off our anomaly detection,
21:13we're actually going to build this again
21:15inside of Claude Co-work. Now, you can
21:18absolutely build this in Claude Code if
21:20that's where you're more comfortable. We
21:22have these built in both for all of the
21:24brands that we work with. Just in this
21:26case, it's really easy to build this
21:28within Co-work, especially if you're not
21:29totally familiar, because it's going to
21:31act a little bit more in a user-friendly
21:33way. So, we're going to prompt something
21:34pretty straightforward. Using the
21:36Facebook Ads MCP connector, I want you
21:38to build an anomaly detector. This
21:40should account for any significant
21:43change over two standard deviations
21:46across the brand that we are looking at.
21:48The purpose of the anomaly detector is
21:51to detect both good and bad anomalies in
21:54my ad account. I would ideally like this
21:56to be run four times per day, and
21:59specifically, we're looking at core
22:01metrics like spend, ROAS, CPA. Oh, we
22:05also care about secondary metrics like
22:07spikes in CPMs.
22:10Okay, now this is a case where I
22:11definitely want this to be built into my
22:13Slack and DM. I want to get this
22:15notification. I want to be warned when
22:17something is actually happening that's
22:19bad. The lookback window when we should
22:21calculate the baseline, this is going to
22:23be interesting. We want to know about
22:25daily changes, but also reference 7-day,
22:293-day, and 14-day changes.
22:33At what level should anomalies be
22:34detected? They should be detected
22:36probably just at the account level would
22:39be most critical, but we'll also go
22:40campaign level. What ad accounts should
22:42be monitored? And in good fashion here,
22:44it could not find the anomaly detection
22:47channel, which is because I haven't
22:48created one for this particular anomaly
22:51alert. So, I'm just going to have it DM
22:52it directly to me. Should anomalies be
22:54flagged for significance? Yep. And then
22:56again, in good fashion, it is
22:58automatically requesting to schedule the
23:00task, so we'll go ahead and schedule
23:02that. In this case, we're running it at
23:036:00 a.m., 11:00 a.m., 3:00 p.m., and
23:057:00 p.m. Really just to call out any
23:07anomalies possible. Okay, so now that we
23:09pull into Slack, we can see this is the
23:11message I got. We have the anomaly
23:13detection, we have the account ID, the
23:15target date, and we have obviously just
23:17a sample here of all the metrics,
23:19yesterday what happened, any deltas that
23:21we've put together. Now, I don't love
23:23this exact anomaly detection. There's
23:26way better versions of this that we can
23:28create, but I just want to give you the
23:30gist of this is what you can actually
23:32create. It is so easy at this point to
23:35just highlight this, take a screenshot,
23:38and get a different version of it that's
23:39way easier for you to actually digest.
23:42Okay, now we understand the basics of
23:43how to build any anomaly detector. This
23:46is essentially just the point where we
23:48can create any sort of pulse that's an
23:50extra set of eyes on an account to make
23:52sure things don't go haywire. After
23:54this, we start to think about how can we
23:56make good decisions, better decisions to
23:58compound on our ad account or compound
24:00on our entire business with winner
24:02detections. So, winner detections are
24:04going to function just like anomaly
24:06detection, except they're going to go a
24:08layer down at the ad level, and they're
24:10specifically going to help us with
24:11future workflows. So, for our winner
24:14detection, what we're going to do here
24:16is we can actually use the same Facebook
24:19Ads anomaly detector, or we could just
24:21take this exact same prompt that we did
24:23last time, and we could build a new
24:25section inside of Co-work or Claude
24:27Code. Again, I'll say this for the 100th
24:29time, use what you're most comfortable
24:32in. Whether you're using Claude Code,
24:34Claude Co-work, or even a different AI
24:37tool, the actual function of what you're
24:40doing is going to be similar, or if not
24:42exactly the same. Okay, so for this
24:45example, we're going to use the same
24:47chat that we previously were building
24:49in. The reason for this is actually
24:50quite simple. When we're using the same
24:52chat, it has all of the context of what
24:54we've already created. Now, you don't
24:56need to do this all of the time, but in
24:58this case, we're going to be using the
24:59same ad account. We're going to be using
25:01a lot of the same skills that we've
25:02already modified. And if you have gone
25:05back and forth in this same chat to
25:07modify how you'd like things to look,
25:09what things you don't like, what metrics
25:11you'd like, then it's going to have a
25:12little bit more context on what you're
25:13building. So, for this example here,
25:15we're just going to build our second
25:16prompt. I want you to now build a second
25:19anomaly detector. Except this time,
25:21we're going to be building it for
25:22winning ads. So, I specifically want you
25:25to look at the winning ads in my ad
25:27account, which we'll define by any ad
25:29that drives more than 5% of the total
25:32campaign spend and is above my target
25:35return on ad spend. For this example,
25:37let's say my target return on ad spend
25:39is one. As long as the ad is driving
25:41over a 1x. I also want you to rule out
25:44any sale periods which you've previously
25:46looked at before. So, we could just
25:48ignore days completely that have massive
25:50spikes in revenue. And I'm talking like
25:535x over baseline. The purpose of this
25:55anomaly detector is to specifically pull
25:58out the winning ads in the ad account.
26:00So, I want a winning ads report that
26:03pulls out the ad ID and any other
26:05contextual information you can tell me
26:06about the ad. Of course, I want the
26:08baseline stats for this ad, but I'd also
26:10like for you to pull these winning ads
26:13based on 3 days of data, 7 days of data,
26:15and 14 days of data. I'm not looking for
26:18a single day spike. I'm looking for
26:20chunks of data analysis so I can make
26:22informed decisions. Okay, now this
26:25prompt is quite long. Feel free to
26:26adjust it how you see fit. The purpose
26:29of this all in all is that we have a
26:32single prompt that's going to give us a
26:35big dump of information and then we can
26:37narrow it down from there. By the way,
26:39all of the prompts that I'm going
26:40through, way more refined, the exact
26:43artifacts, the exact prompts to build
26:45this all for you is always inside
26:47school.com/facebook. You could join at
26:49any time. You could steal all these
26:51prompts and leave the next day. It
26:52doesn't matter. Okay, how often should
26:54this report run? Once per day. Should
26:56the report go? Yep, same as Slack.
26:58Should we define the sale period day to
27:00exclude? We'll just go here. That's
27:02fine.
27:05Okay, we can see this has finished up.
27:07Let's take a look inside of Slack real
27:09quick. So, we can see this is a massive
27:12data dump that we've gotten here. Now,
27:13obviously this might be too much for
27:15you. It's definitely too much for me
27:16right off the bat, but it does give me a
27:18general idea of what's going on. The
27:20cool thing is I can now curate this to
27:23my specific business. Great example
27:25here. Let's say here we have our
27:26retention campaign up top here. This $50
27:30represents 20% of the total campaign
27:32budget. So, what I would obviously just
27:34do here is I would just take this and
27:37I'd say maybe I want to break this out
27:38by campaign type. Maybe I want to break
27:41out retention versus prospecting. So, I
27:43don't have everything grouped together
27:45here. Maybe I only want certain kinds of
27:47ads like certain kinds of formats of ads
27:50or ads that only have a specific naming
27:52convention in them. Here's the good
27:53thing that we see here. There's this
27:54persistent winners across all three
27:56windows. I think this is pretty cool.
27:58Overall, this is solid for me to
28:00understand generally what's going on.
28:01Here's the miss that we have. We don't
28:03have links. We don't have pictures. We
28:05don't have images. So, everything you
28:07would build on top of this would be to
28:09enhance this data dive. Now, what I
28:11would actually do here is I'd actually
28:13like to see this in two different
28:14places. So, don't get me wrong. It's
28:16nice to have a report like this, but
28:18this is overwhelming, right? I would
28:21have to modify this report pretty
28:22heavily. What I'd actually like here is
28:24to build an artifact around this. So,
28:26we're just going to tell Claude this is
28:27pretty solid. Let's now just build a
28:29persistent artifact around this.
28:31Something that I can go in and reference
28:33at any time. I think that it would also
28:35be helpful to group things by campaign.
28:38It currently looks a little bit
28:39overwhelming in its current state. Keep
28:41in mind that I'm looking for things to
28:43be sorted by amount spent.
28:46So, all we're going to now get with this
28:48is just a slightly tidied up version of
28:50what we've already built. Okay, great
28:52example of how you can build absolutely
28:54nothing of value if you don't prompt
28:57properly. I clearly did not build the
28:59right thing. So, we're just going to ask
29:00Claude to actually just look at this for
29:02us. Okay, it looks like something has
29:04went wrong here. I've got no information
29:05here. This is completely different than
29:08the Slack report that you sent to me.
29:10So, I'd like you to take screenshots of
29:12your own artifact that you built here
29:14and try to figure out what went wrong,
29:16troubleshoot it, and create a new one.
29:18And just like that, we now have our
29:20winning ads report. Again, modify this
29:23to your liking with the metrics that
29:24matter the most for your business. Now,
29:27I think some things that I would
29:28probably add right in here right away
29:30would be things like links to the actual
29:32ads. If we could eventually pull images,
29:34that would be huge. But really, just one
29:36of the easiest things we can do is
29:37instead of having us to then look for
29:39what is this ad, can we then take this
29:42ad and click a single link to then drive
29:44us right to the ads manager to preview
29:46it. Right? These are just easy things
29:48that we can do that allow the entire
29:51process to be much easier for ourselves.
29:53Okay, so instead of tweaking this a
29:54hundred times over now, we're just going
29:56to keep this as is. Feel free to modify
29:59to your liking. The most important part
30:01of everything that we're about to hit is
30:03going to be what happens after. All
30:04we've set up to this point is our basic
30:07setup, our basic capabilities of Claude
30:09in the Facebook Ads MCP, how to set up
30:11and schedule reports, and how to build
30:14both anomaly detectors for negative
30:16situations and winning ads. What we do
30:19moving forward is literally tying it all
30:21together, and this is how it becomes the
30:24agent for your business. This is how you
30:26actually grow your business fast and
30:28actually produce more for your business
30:31using Claude's code and the Facebook Ads
30:33MCP. So, now moving on to number seven,
30:36connecting additional business
30:37information. This is the context that
30:40that going to power all of the outputs.
30:4299% of people will have already stopped
30:45watching this course at this point, so
30:46consider yourself in the 1% that's going
30:49to have a major advantage. At this
30:51stage, our goal is to give Claude as
30:53much information and context as
30:56possible. This is going to deliver
30:58significantly stronger outputs for
31:00creative workflows and any other
31:02workflows that we build after. So, for
31:05number seven here, I'm going to load in
31:07two very important pieces. The first
31:10thing that I'm going to put in is going
31:11to be the brand guidelines, and then the
31:13second thing we're going to put in is
31:15the output of our avatar building
31:17prompt. Brand guidelines and avatar
31:19building prompts, the exact ones that I
31:20use are always in our school community
31:22updated based on what's actually working
31:24the best right now, but you could
31:25probably put these together on your own.
31:27So, to do this, we're going to build a
31:29very basic skill inside of Claude. We're
31:31going to click a new task here, and then
31:33we're going to say, "Build a skill that
31:35you could easily reference." This is
31:37going to contain my brand guidelines and
31:40my avatars for my brand. Then, we're
31:42going to dump all the information, so
31:44this is going to be the brand guidelines
31:46first. Then, we're going to go into
31:48another tab right here. We're going to
31:49grab this avatar building sheet, just
31:51make a quick copy of it. We usually dump
31:53our brand guidelines right here per the
31:55SOP. We drop this into a new context of
31:58Claude, and we're just going to give
32:00this in the highest model available.
32:02We're going to drop this in. This goes
32:03through a very extensive research
32:05process. This is something that we've
32:07built for a while. It works tremendously
32:09well. Usually, this takes a few minutes
32:11to run. So, as this is working, just so
32:13everybody understands what this entire
32:14prompt does, it basically scrapes the
32:17entire internet, specifically forums and
32:19Reddit, to get critical response and
32:21drive customer avatar clusters. This is
32:24really how marketing works in 2026 and
32:27onwards, which is where you specifically
32:29talk to an avatar. You build problem and
32:32solution-oriented creative for that
32:34avatar. And you can kind of see as this
32:36is still going here, we have stage one,
32:38which are 20 problems in the category.
32:39So, it identified the category, it found
32:41a bunch of competitors, and then it
32:43found specific problems in this category
32:45that are all over. And you could see all
32:47the different places Useful Vitamins,
32:49eBay, Goodreads, Fortune, Grooons, Team
32:51Blind, Substack, all these different
32:53places that this information and
32:55problems are pulled from. And then we
32:56have 20 solution angles for the brand
32:59specifically. We're just using Grooons
33:00here for example. And then it's driving
33:02us different avatars that we actually
33:04have for the brand as well. So, it's
33:05putting together all the avatars for us.
33:07And again, you want to, you know, refine
33:10this to your brand for what actually
33:11fits to you. For this example, I'm just
33:13going to use this. Honestly, this gets
33:15us like 98% of the way there. Usually,
33:17it's just some tweaking. Maybe we remove
33:19one or two avatars in total, but we
33:21usually get really, really good clusters
33:23with just the two prompts that I showed
33:26you right there. And this saves probably
33:272 days of work. Also, again, you could
33:29always get it inside the community.
33:31Okay, so I'm just going to copy all of
33:33this from the clustering start. We're
33:36then going to drop this into our code
33:38instance here, and then we're just going
33:39to write avatars.
33:41I should probably understand what I'm
33:42saying here. So, this is a lot of
33:44information. I'm going to make sure
33:45we're using Opus 4.8 high here because
33:48this is going to be pretty critical. And
33:50we just want this to write the skill for
33:53us because then this skill is going to
33:55be referenced later. Again, I'll say
33:57this for the 100th time, this could be
33:59done in a hundred different ways, and
34:00there's probably a more efficient way to
34:02do it. But, I'm just showing you the
34:04easy first step that I would take,
34:06especially if I wasn't 100% comfortable
34:08in here. So, we're just going to easily
34:10just move this through. We're going to
34:12go ahead and send it.
34:14Great. Now that this is done, this only
34:16took a few minutes here. We actually
34:17have the full skill. This is just a
34:19referenceable skill. You could do this
34:21in a few different ways, but we're just
34:23going to go ahead and save this skill.
34:25We can now reference it at any time, or
34:26we could just tell Claude to reference
34:29the skill. Meaning, next time we ask for
34:31copy, or creative or what we should spin
34:34up next, which we're going to do in a
34:35few minutes here, we're actually going
34:37to just be able to say pull up the Gruns
34:39brand skill at any time and it will
34:41reference it. In fact, as we continue to
34:43use the same thread, which we've done a
34:46few times now, it's going to naturally
34:48reference it because it's smart enough
34:49to understand that the skill applies to
34:51the business. So, all of that done here,
34:53we've now connected additional business
34:55information that's going to help with 8,
34:579, and 10. This is going to be basically
35:00the groundwork that you could build upon
35:02any time. And by the way, you could add
35:04more information, upgrade your skills at
35:06any time, and it's so easy because all
35:08you need to do is just tell Claude to
35:09upgrade, change, or modify the skill.
35:11Moving on to the next piece here, we
35:13have creative recommendations. This is
35:16going to function very similar to our
35:18winning detection. The only difference
35:20here is we're now going to start to
35:21think about this from a workflow
35:23perspective. And we're going to be
35:24adding a lot more context than we had
35:26previously. I'm just going to ask it,
35:28"Can you pull the top spending ad for me
35:30with the ad ID and link it in Ads
35:32Manager?" Also, can you pull the actual
35:34creative so I can see it? Here's the
35:36great thing. Whenever you ask for
35:37something like, "Can I see it?" that
35:39also should mean in most cases that
35:41Claude can see it, too. We have the Ads
35:43Manager click, so we can click this. We
35:45can then open this up right away inside
35:47of Ads Manager, which is great. We then
35:49also have the live ad preview and then
35:51the video that we can watch on Facebook.
35:53What's really cool about this is at the
35:54bottom here, it noted, "The image above
35:56is a rendered creative, a vertical
35:58before and after split showing XYZ XYZ
36:01XYZ." This is excellent. We know that it
36:03could actually see the ad for us.
36:05Because of this, we now want to use this
36:07workflow to then start to generate more.
36:11This is where things get fun. So, here's
36:13what we're going to build, and it's very
36:15easy. Please build a basic schedule that
36:19pulls exactly this information that you
36:21have above, taking my top spending ads
36:24over the last 14 days. Do this schedule
36:27only once per week and focus on the top
36:30three highest spending ads in the
36:33account. Ignore everything else.
36:44So, all we're building again is a basic
36:47schedule. You're going to see how this
36:48all starts to tie in together here. Now,
36:50while that's working in the background
36:52here, I want to explain the difference
36:53between winners and potential ads. It's
36:56very, very simple. You just need to
36:57define your winning parameters based on
37:00your efficiency metric and as long as
37:02that ad has spent. Typically, we like to
37:04think of winning ads as having a hit
37:07rate. As long as both of these
37:08conditions are met, then it is in fact a
37:11winner. Potential in our case would be
37:13something like 5% of campaign budget,
37:16but missing TROAS or TCPA. Or on the
37:19flip side, hitting TROAS or TCPA, but
37:22less than 5% of the budget. Potential
37:24ads need to be tweaked. Winning ads need
37:27to be replicated. So, the workflow that
37:29we're going to now start to create is
37:31going to be based on these winning and
37:33potential parameters. Keep in mind, I'm
37:36not narrowing down all the edge cases
37:38for every single business. Just in your
37:40case, as long as you're hitting your
37:42specific goals and targets, that's what
37:45you want to build your workflows around.
37:47Back into Claude here, we have our three
37:49ads that we pulled through. From this
37:51stage, we're now going to build on top
37:54of this one last time that's going to
37:56start auto-generating work for us. All
37:59I'm going to say here, again, in plain
38:01English, for any of these top ads that
38:04are image ads, I want you to take the
38:07image preview and build a variety of
38:11dynamic iterations of the same ad. Feel
38:14free to change any component that you
38:16want that you think would be most
38:18important. Don't forget to reference our
38:20brand guidelines skill that we built
38:23together earlier. Every time you create
38:26these new images, I want you to review
38:29them before you send them off to me to
38:31make sure they follow the brand
38:33guidelines avatars, and feel free in a
38:36few of these variations to be unique and
38:38creative. We're not looking for tiny
38:40little tweaks or iterations, we're
38:42looking for moderately big changes based
38:44on an original ad that we already know
38:47drives a lot of spent.
38:50Okay. So, the point of this now is as
38:52you can probably imagine, we're building
38:54workflows. So, our creative
38:56recommendations, right? This is really
38:57easy. This is just building the
38:59foundation of what's going to be
39:01recommended and what's going to be
39:02created in our workflows. Creative
39:04workflows is really where the fun
39:06begins. To put this into perspective,
39:08let's say we have all our winning ads up
39:10top here. What we want to do is every
39:12time there's a winning static ad, we
39:15want to create more iterations. So, we
39:16have the V1, we have the V2, we have the
39:19V3, and so on. And what winds up
39:20happening is this grouping of ads
39:24becomes stronger and stronger with the
39:26winner. Because we have a positive
39:27signal based on high spend and hitting
39:30our target, and it actually expands the
39:31entire ad account, and makes the ad
39:33account more bulletproof. You'll see
39:34less fatigue in the account. It's a very
39:36positive thing that happens. In addition
39:38to our winners, we're also going to have
39:39our potential ads, which we'll call
39:41right here. Now, for our potential ads,
39:43we're going to do the same thing. We're
39:44going to create a V1, a V2, and a V3,
39:47right? And every time we do this, every
39:48time we're creating these new versions,
39:49especially for potential ads, we're
39:51looking for variety. We're not trying to
39:54create the same ad over and over. We're
39:56trying to create different ads. What
39:58you've basically done by just doing this
39:59workflow right here is you have
40:01massively expanded your ad account. You
40:03now have creatives that are flying in on
40:05a regular basis. You're giving these
40:07creatives 14 days to run. You're
40:09analyzing these creatives right here,
40:11and then guess what you're doing? You
40:13are creating more creatives, right? So,
40:15you're constantly building this natural
40:17creative flywheel with your winners and
40:20with your potential ads. Okay, so after
40:22you run this, you're probably going to
40:23notice you run into a big limitation.
40:26That limitation is going to happen
40:27because you have all this context, you
40:30have all of these briefs, but Claude
40:32cannot generate images for you
40:34automatically. It could do so in regular
40:36chat. These images are often very weak.
40:38Claude is not the best image generation
40:40tool. ChatGPT tends to be a bit
40:42stronger. Nano Banana tends to be very
40:43strong as well. Now, this is the point
40:45where you get to customize your
40:47experience based on what you and your
40:49team actually like to use. For example,
40:51my team uses Super Scale primarily. We
40:53used Higgs Field a little bit as well.
40:55And then for static images, we tend to
40:57use Nano Banana Pro. We tend to use Arc
40:59Ads as well. All of these tools can be
41:02used at your own discretion. We are a
41:04large agency. We work with a lot of
41:06brands, and because of that, we have the
41:08capability to use everything under the
41:10sun based on what's the right fit. So,
41:12at this point with the briefs that we've
41:14created here on the right side, we kind
41:16of have an idea of where things can go.
41:18But, as I've mentioned before, we've hit
41:20a clear wall. We're not actually
41:22generating anything, and that's just
41:23because Claude itself is not very good
41:26at generating images or generating
41:28videos. There's some license to do this,
41:30but not too great. There's a whole lot
41:32of other tools that we can use that are
41:33much better. Think of just Claude as
41:35like the brain, the architect for all of
41:37this, and what we're going to use is
41:39other model context protocols, other
41:41MCPs, to tie this all together. So,
41:44keeping this as simple as possible, I'm
41:45going to show you how we set this up in
41:47Super Scale because this is who we use.
41:49So, if you click on Super Scale down
41:51here, where the MCP is, and you
41:53literally just copy this MCP. You can
41:55then go into Claude. You can then go to
41:58your settings in Claude. You can go to
42:00connectors. Click customize right here.
42:02Click the plus button. Add a custom
42:04connector. We'll just call this Super
42:05Scale, and we'll just paste this in. And
42:07once we just click this add button right
42:09here, we're going to have access to
42:11interact with these Super Scale MCP, aka
42:14create images, videos with context using
42:17the brain that we've already created to
42:19actually generate things for us on
42:20repeat. So, I'll just click add. So,
42:22then if we go back to our chat here, we
42:23can prompt in something very simple.
42:26Okay, this is fine. Now, since you can't
42:28generate images on your own, please use
42:30the MCP to build these images for me.
42:34You can go back in the chat, take one of
42:36these prompts, and just build one image
42:38ad for me. I want to use Nano Banana Pro
42:41and build two variations per iteration
42:44that's necessary. Do this all in
42:46vertical formats.
42:53Now, just to show you what's going on
42:54here, we're using Higgs Field for this
42:56job. I could actually specifically tell
42:58it to use Super Scale. I could tell it
43:00to use Higgs Field. I could tell it to
43:01use anything that I've set up. It
43:03doesn't really matter at all. As I said,
43:05I tend to get the best results from
43:07Super Scale, but this is going to just
43:08go really quick. Now, you can see these
43:09images starting to be generated on the
43:11fly here, and just like any AI-generated
43:14image, it's going to pull in in just a
43:16few minutes. Now, really I want us to
43:18understand the workflow of what we just
43:19created here because this is critical.
43:22So, first off, every single time you
43:23have a winning ad in your ad account, we
43:25are creating somewhere around two to
43:27four briefs. These briefs are going to
43:29have context on who our avatars are, and
43:31they're going to be able to look at
43:32historic winners in the ad account, the
43:34images specifically. They're really
43:36going to be able to understand. And then
43:37we're going to pull in those briefs into
43:40either Higgs Field, Super Scale, Arc
43:42Ads, Gemini, whatever tool you like to
43:45use to generate more images. You have
43:47essentially created, as we went over
43:49already, a key creative flywheel. Your
43:53images are now going to be able to
43:54repeat and repeat and repeat based on
43:57what's actually working, and you could
43:58run this technically anytime, although
44:00I'd recommend 7 to 14 days on a
44:01consistent cycle. Okay, so now we've
44:03created an auto-generating image machine
44:06that you could basically prompt on tap
44:08whenever you want and auto schedule it.
44:10What about scripts and briefs? Now,
44:12scripts and briefs are going to be
44:13something that you actually do more on
44:15demand, not necessarily always on a
44:17schedule, but we get also create a
44:19similar workflow for both scripts and
44:21briefs. Now, what I like to do for all
44:23of the briefs that we follow is I like
44:24to create this as a skill as well. So, I
44:26have our UGC creator brief template
44:29here. This is for internal and external.
44:32We like to basically push this all into
44:33one spot so that we get the optimal
44:36output. And even though most of this is
44:38blurred because it's only available
44:39inside of the Facebook Ads Blueprint,
44:42just to give you like one example, we
44:44are crafting individual hooks, we're
44:46crafting individual call-outs, and we're
44:48preparing additional shot lists so that
44:51when we actually get the creative handed
44:53to us, whether it's AI creative, UGC, or
44:56in-house product studio creative, we
44:58know that we're hitting all the marks.
45:00Briefing is critically important. If it
45:01wasn't, everybody wouldn't talk about
45:03it. So, what we like to do is we like to
45:05take this brief just like so. Then we're
45:07going to pop this back into Claude. And
45:10what we're going to run is very simple.
45:12I've just prompted here, build a skill
45:13that could easily be referenced for
45:15scripting video ads. I've then provided
45:17the full template, and then the skill
45:19was then created. Now, anytime I want
45:21to, we could edit this skill, we could
45:23reference this skill, and anytime I want
45:25to create a new script, of course, it
45:28will automatically reference it. For
45:29example, let's say I said, "Using the
45:31Facebook Ads MCP in the account, please
45:33find my top-performing video ad, extract
45:37the transcript, and then using the skill
45:41that you've just made, write a new
45:43script and brief."
45:47So, just like all other workflows that
45:49we've created, you know what's going to
45:50happen now. It's finding the transcript,
45:53it's doing exactly what we did before,
45:55and then it's going to brief in exactly
45:57what we want in our own template that we
45:59can pass to our internal team, to our
46:02designers, to our creators, whomever we
46:04want. This ultimately saves you hours
46:06and hours over time and time and time
46:09and time again that you have to manually
46:10create over and over briefs and briefs
46:13and briefs again. So, at this point you
46:14understand that we've created basic
46:16creative workflows, generating images,
46:19building scripts, building briefs, all
46:21of which could either be auto-generated
46:24for us depending on what tools we're
46:25using, or we can actually just use these
46:27to our advantage to brief creators, and
46:30so on. This is generally the point of
46:32where things are good right now. But, by
46:34the time this video releases, there's
46:36going to be a hundred new things that
46:38come out as there always are. And
46:41because of that, I want you future-proof
46:43where your mindset goes on this because
46:46every week when new advancements come
46:48out, there's going to be more that we
46:50can do. This is the foundation for
46:52everything that you're going to do in
46:53the future. Future workflows are
46:55essentially going to be limitless, but
46:57what I really want you to focus on are
46:59just a few things. So, the core workflow
47:01that we created is finding winning ads,
47:03briefing winning ads, and then of course
47:05using some tool to build more of these
47:09ads. This is all currently done pretty
47:11well with image ads, right? We'd all
47:13agree that right now AI image ads are
47:15pretty great, but when it comes to
47:17video, it's a little iffy. So, what we
47:19like to do from this point forward in
47:21the future workflows that I think are
47:22going to be extremely valuable are how
47:24we take one single brief and multiple
47:27tools to then create video iterations
47:31and image ads all together. So, instead
47:33of just thinking of this as an image ad
47:35briefing, we would potentially take a
47:37winning ad, whether it's an image or a
47:40video, we would then brief into the
47:42tool, but we'd also brief images, and
47:44we'd also brief video, right? And again,
47:46this is just the start of things. So,
47:49maybe we'd have one tool for images and
47:50a different tool for videos. It doesn't
47:52matter. Really, we just want to make
47:54sure on autopilot we're getting our
47:56winning ads, we're briefing in images
47:58and videos. And the cool thing is you're
47:59going to be able to modify the briefs
48:01based on what's necessary for the tool
48:03or for the concept. If it's an image
48:06concept, we know it could be simple,
48:08iterations are really easy to make. If
48:10it's a video concept, then we might need
48:11to pre-prompt, and those are all done
48:14with skills. So, just think of skills as
48:17your intermediary between right here.
48:20So, skills are not the answer to all,
48:22but they're going to be the bridge in
48:24between the creation of the brief and
48:26what you actually get on the output. So,
48:28a skill would typically sit right here.
48:31Now, a skill in many cases is going to
48:33just be this entire workflow, but the
48:35way that we're using skills here is to
48:37interfere and be a buffer between the
48:40brief and the actual output. What this
48:42saves for you is overspending on token
48:46cost, because spending too much on
48:47tokens is going to cost you a fortune in
48:50the long haul, especially if you're
48:51doing this on repeat. If your skills are
48:53better at filtering, you're going to
48:55save on tokens, you're going to produce
48:58better ads with higher hit rates. So,
49:00future workflows that I think are going
49:01to be great are going to be video ads,
49:03custom scripts per platform or tools,
49:05and then I think there's going to be an
49:06entire suite that we eventually layer
49:08into all of this with finance and
49:10inventory. Now, that is TBD at the
49:13moment, there's not as many great
49:14connectors as we would like here.
49:15Shopify does port in, but the
49:18capabilities right now are a bit limited
49:20in terms of the financial role of
49:22everything. Ideally, in the long term,
49:23we create a full ecosystem that allows
49:26us to take the most basic reporting,
49:28integrate that with our creative
49:29workflow, that eventually spits back out
49:32to the finances and the inventory of the
49:33business. On a long-term scale, we're
49:35essentially creating a massive flywheel
49:38in the business that has dozens and
49:40dozens of different components. So,
49:42while you might have your creative
49:43flywheel, for example, you could also
49:45have additional layers in here, for
49:47example, making sure that only this ad
49:50is evaluated if that ad is truly
49:52profitable based on not only the
49:55attribution of the platform, but
49:56profitable based on the unit cost or the
49:58cost of goods or the shipping threshold
50:01or the inventory available. Right? So,
50:03you could see how we're going to
50:04eventually be able to layer on over and
50:06over and over again different things
50:08into the business to make the
50:09decision-making even better. Overall,
50:12the workflow that we've created today is
50:14the foundation and then what you build
50:16on top of this should be custom to your
50:18own business. Okay, before we wrap up, I
50:21want to hit you with some of the
50:22biggest, most critical limitations. Now,
50:25you might be thinking of limitations as
50:27something that is not available today
50:28and it's going to be available tomorrow.
50:30That is one way to think about it, but
50:31think of this more as a red flag. Some
50:34people are going to run with Claude in
50:35the Facebook Ads integration, upload all
50:37their ads tomorrow, they're going to
50:38create massive flywheels, and their ad
50:40accounts are going to get bloated really
50:41quickly. They're going to have a bunch
50:42of crappy ads that don't perform or they
50:45might have a few good ads, but they
50:46never get adequate spend. And this is
50:48going to be a death loop for most
50:51businesses. Tools like this are
50:52fantastic
50:54at enhancing what you're currently
50:56doing. They will make you more efficient
50:58if done right, and if you build your
51:00workflows custom to yourself, they will
51:02really enhance your entire organization.
51:04So, what we do here at The Moon Liters
51:06is we use everything that I just showed
51:08you as a filter, almost like a step one
51:11into just the foundation of the critical
51:14mass or media buying is still very, very
51:16relevant. The most important thing to
51:18understand is that at the end of the day
51:19when it comes to running Facebook Ads,
51:21you're spending thousands and thousands
51:23of dollars, likely of your very own
51:25money, you need to make sure you have a
51:27human filter at the end of that. Just
51:30because everything looks sexy and shiny
51:32right now does not mean that it's the
51:34appropriate thing to do for the business
51:36today. So, be careful, limit yourself,
51:39don't overdo it in your ad account, and
51:41most importantly, do not remove the most
51:44important filter from making sure your
51:46cash is used properly. That is
51:49everything that I'm going through today
51:51in the Claude Meta Ads MCP tutorial.
51:54This is the start of a new era, and if
51:57you want to see everything I just went
51:58through, not only in this format, but
52:00the exact prompts, all of the briefs I
52:02showed you, everything laid out in a
52:04very easy step-by-step version, then
52:07we've added all of this to
52:08school.com/facebook.
52:11You can join there anytime, or if you
52:13want all this handled for you, if you
52:14don't want to ever have to deal with
52:15this, you can go to the
52:16moonlighters.com/apply
52:18to see if you're a fit to work with us.
52:20That's all for today. I hope you got a
52:21ton of value out of this, and I'll see
52:22you in the next one.

About this transcript

This page contains the full transcript of Claude + Facebook Ads (FULL COURSE) by Sam Piliero, generated from the public captions YouTube serves with the video. The transcript has 10,772 words across 1,536 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.