Claude + Google Ads

Micah Johnson built a fully functional Google Ads lead system using Claude, and it's generating real leads without a specialist or an agency involved. In this episode, him and Alane Boyd walk through every layer of the build: campaign research, ad copy, conversion tracking, the CSV export workflow, and the daily AI-powered review that keeps it running.

If you've ever tried to run paid ads and gotten stuck in the weeds of conversion tracking, keyword strategy, or Google Analytics event setup, you know how painful it is. Most businesses either give up or hand it off to an expensive agency. Alane and Micah found a third way, and the results are already coming in.

In this episode, you'll learn:

• How to use Claude to research and build a full Google Ads campaign from scratch, no specialist required
• The MCP-plus-CSV workflow that lets Claude do the heavy lifting on Google Ads without touching the account directly
• Why heat map data mattered so much, and how Microsoft Clarity revealed visitors were copying their answers and leaving instead of converting
• How they qualified leads at scale with Claude, and what 300+ leads taught them about a blind spot in their own targeting
• The daily routine that pulls Google Ads, Analytics, heat map, and lead data into one place for a single recommendation
• How the leadership team stays in the loop with zero manual reporting, through a Slack channel and a shared Cowork dashboard

🛠️ Tools & Platforms Mentioned

• Claude: AI used for campaign research, ad copy, tracking setup, and daily analysis
• Claude Code: Used for writing JavaScript event tracking code and tagging
• Claude Cowork: Used for shared leadership dashboards and presenting campaign results
• Google Ads: The paid advertising platform used for the campaigns
• Google Ads Desktop App: Used to upload Claude-generated CSV files and push live
• Google Analytics: Used for tracking events and conversion data
• Microsoft Clarity: Heat mapping tool used to analyze landing page behavior (rage clicks, copy-paste behavior)
• Slack: Used for daily lead and campaign analysis digest for the leadership team
• MCP (Model Context Protocol): Used to give Claude read-only access to Google Ads data

Learn More about our AI Leadership Workshop

Winning businesses aren’t just working harder, they understand how to use AI strategically. Sign up for the AI Leadership Workshop, a 3-hour expert-led mastermind designed to give leadership teams clarity, alignment, and a practical path to stay competitive.

Book a call to customize your workshop today!

This episode is brought to you by Biggest Goal.

Every quarter your team spends evaluating AI is a quarter your competitors spend shipping. Most leaders feel the pressure but get stuck between ignoring AI and getting it wrong. More tools and more demos won't fix it. What actually works is hands-on training for the people doing the work.

For more information about our AI education programs, visit biggestgoal.ai.

Free Resources & Upcoming Sessions:

• Free 30-minute Claude Cowork Demo
• Claude Cowork Masterclass
• Asana, ClickUp, or Monday.com Selector Tool
• 25 Free Custom Automation Ideas for your Business

Connect with Us:

• TikTok
• LinkedIn
• YouTube
• Community

Disclosure: Some links are affiliate links, meaning we may earn a commission at no extra cost to you. Thanks for supporting the podcast!

Transcript

Alane Boyd (00:00)
Hi there, I'm Alane Boyd.

Micah (00:03)
And I'm Micah Johnson.

Alane Boyd (00:05)
And you're listening to Automate Your Agency. Ooh, today we are going to be talking about the Google Ads Lead system that we created and built with Claude Cowork, and it's absolutely incredible. I'm so excited to share this with you.

All right, Micah. So this was something that you and the team built without me being a part of it. I knew that y'all were working on it, but I didn't see the ins and outs of it. But I get to be the recipient of the information, seeing the updates every day in our daily review of the ads campaign. but what was the very first thing? Like I don't know how you did this. I just know you used Cowork to help build out Google Ads. What was the first thing that you did?

Micah (00:49)
I mean, for this use case, the first thing that we did was follow our start small methodology. And that is literally just get down to the basics and provide Claude with enough information on what we wanted to create an ad campaign about and just start there. So we worked with Claude, we showed it our offers, we showed it some of our messaging, we gave it some of our design style, we showed what works.

For some of the stuff that we've done in the past. And then we asked it to shape this ad campaign for us. And essentially it was just it putting together and throwing out ideas. We had it do

Alane Boyd (01:31)
Mm-hmm.

Micah (01:31)
some research online, which helped a lot. And you know, but there's it was a human element every step of the way. So we would look at what Claude was saying and then come back and go, Yeah, let's steer it this direction or let's go this way. It wasn't

Claude built everything and then we just pushed it live and, you know, hoped and prayed that it worked well. Like it it took Claude's research, Claude's knowledge of the paid ads for Google in particular that we're talking about right now. It's an expert on that, which is great. And we're

Alane Boyd (02:06)
Mm-hmm.

Micah (02:06)
the expert on what we're doing. And so we merged those two together.

Alane Boyd (02:10)
Yeah, I know in the past, we've tried to get ads off the ground, or we did, we got ads off the ground. But even just the time to do that research piece on, what would be good keywords or what would be good landing page ideas or you know, where are we sending them? Like it did all of that research for us. And for us to do that, it was so time consuming. I know that's a big word that we use when we're talking about AI, but it is.

And we may not get it right. We're based off of just where we are at our time trying to do that research, trying to look at particular other companies and model it off of them or whatever it might be. We just couldn't really put together the bigger picture trying to operate as ourselves.

Micah (02:52)
Yeah, and I would say there's a layer to this that surprised me as we went through this. and I'll get to that layer here in a second, but to kind of answer your last question, We gave Claude MCP access to Google Ads. The catch with that is it's read only.

So Claude can't go and just create all the campaigns and create the ad groups and manage all the budget and do all of that. There's some platforms that you can pay for, but I also we didn't want to give Claude that much power. I like the idea of saying, hey, let's have a conversation with Claude, who can act as an expert and kind of an outside perspective with different context, and then be able to produce assets that we can then use and the work around or the path that we've

successfully followed is we have Claude export CSV files and then we downloaded the Google Ads app that you can install on your desktop. We take those CSV files that are already perfectly formatted from Claude to do all that heavy lifting, upload that to the desktop app, and then we can preview everything. And when we

Alane Boyd (03:56)
Mm.

Micah (03:57)
are good with how everything looks, we hit post and that pushes it to the ad campaign that we have. So

Very, very like there's no copy and paste, there's no like tedious stuff. The extra layer that I was talking about a second ago really surprised me, which is one of the things that we had messed up in all of our previous attempts for paid ads, Alane, was the nuance of the campaigns. So all the different settings, tracking conversions,

Alane Boyd (04:31)
Mm-hmm.

Micah (04:33)
tagging,

triggering Google Analytics events, getting all of those things to work correctly is such a pain in the ass. It just it's so

Alane Boyd (04:42)
Mm-hmm.

Micah (04:43)
time consuming. It's so complicated. And the first time you do it, you're like, I don't am I doing this? Like which option do I like targeting?

Alane Boyd (04:51)
Mm-hmm.

Micah (04:52)
Or another really good example is the ad strategy. Well I want conversions or I want this as the end result, but

You can't start a brand new campaign and go right to like track conversions on a brand new campaign

Alane Boyd (05:07)
Mm.

Micah (05:07)
with a brand new landing page with all this brand new information because Google's algorithm doesn't have enough to know or you don't even know

Alane Boyd (05:14)
Mm-hmm.

Micah (05:15)
if you're gonna get conversions with this messaging yet. So

Alane Boyd (05:19)
Yeah.

Micah (05:19)
like Claude was so helpful. We could dig into all of those, but like Claude was so

Alane Boyd (05:23)
Mm-hmm.

Micah (05:24)
helpful to shorten the cycle to get that stuff answered and launched and tested. I

Alane Boyd (05:30)
Mm-hmm.

Micah (05:31)
I don't

Faster when you've ever done and this campaign is working better than we've ever had anything work.

Alane Boyd (05:39)
Yeah.

So with the other platforms or even just like the Google Analytics pieces, so how much of it was you having to go and do those things manually and how much of it was working through co okay, and through Cowork or code?

Micah (05:52)
Mostly with this, it was code because we're dealing with the websites, we're dealing with tagging, we're dealing with JavaScript for triggering events. So we want code to help us write all of those things. And I would say I I don't know if we used Cowork too much. I think Cowork was like some original ideas, and I think presenting back some of the stuff that was accomplished.

We leverage Cowork to do that work, to present

Alane Boyd (06:21)
Mm.

Micah (06:21)
to our leadership team, but code for the building for sure.

Alane Boyd (06:26)
Okay. And when we say code too for our listeners, we're referring to Claude Code, just so that we're we're being clear. So so yeah, you're not

Micah (06:31)
Yes, not just random code that we're writing, yeah.

Alane Boyd (06:34)
just over there writing a bunch of code. You have assistance. so then you got everything launched with the correct tracking, and we had campaigns running. And what I got to see is I know there's another part to this that we're not gonna get in today, but I think we could do another episode on is the lead scoring.

and how

Micah (06:52)
Mm.

Alane Boyd (06:53)
we were tracking those conversions. So we got to see these in real time, these leads converting. But you also set up a second part to it. And if Micah, if I'm missing something, tell me. But the other side of this is okay, we've got it set up and we've got it working. We see leads coming in, but we need to also review what we're doing and see, hey, do we need to be changing some of the keywords? What keywords aren't performing well?

what keywords are giving us the best results and start analyzing those things. And that's the other side of this that you all built, that we could get a digest essentially of the campaign, how it's doing.

Micah (07:34)
Yeah. Yeah. And I think one thing that I wanna point out in this is, you know, we hear this from clients, Alane, and y it's hard not to be guilty of this where you look at a system like this and you're like, you paint this picture in your mind of let's have Claude every day check all this data and these stats and these metrics and the leads and and then make tweaks for us.

And every day it's gonna learn and get better and better and better. And we don't have to do anything. Claude will just, you know, or AI will just make all these changes. We have to avoid that. That is not reality. Right?

Alane Boyd (08:17)
right. You have the humans in there taking and making the judgment calls, making the decisions, figuring out, okay, if if it's not working, what are our ideas to improve it?

Micah (08:30)
That's right, that's right. Cause AI is only as good as the context, and we can give it lots of metrics and we can give it lots of data, but it doesn't have the same amount of context that we have in our heads of understanding where we're going, where we want to go as a business, what actually works for our audience, etc. And so what we want to do with AI is automate the hard parts, the heavy lifting, the time-consuming stuff. So

In the beginning, setting up the campaigns, it's well, how should we track these conversions? Are they primary or secondary? How do we assign the values? How do we track the events? Are they key events in Google Analytics? How do we create the reports so that we can see what's going? Like all of that is so unknown to most, especially business owners and founders. That's why you have to hire exclusive teams that are specialized in this. But Claude is

Alane Boyd (09:26)
Mm-hmm.

Micah (09:27)
Quite good at getting this off the ground. And so even just writing the event code. But all of that is Claude proposing, yo, this is what I think we should do, and humans approving. And then Claude going, Great, got it. That's even better than what I proposed. Let's make these tweaks. Let's get it in there and initiate all of that. So Claude helped us launch

All of that stuff. That was fantastic. Now the review piece in in between these two, we've got leads coming in. We've got a qualification piece, which leverages AI, we've got a database store, and we've got our CRM that lead information is getting logged to. and then where this kind of ends up is we have Claude on a daily basis run a routine that says, hey, let's check the landing page heat map data.

Let's check the Google Analytics data. And because in the beginning Claude helped us set all this up correctly, we now have

Alane Boyd (10:27)
Mm-hmm.

Micah (10:28)
like the events in Google Analytics. So Claude actually has better context than just, hey, how many visitors and for how long? All those little nuances

Alane Boyd (10:37)
Mm-hmm.

Micah (10:38)
combined with the heat map data, combined with the MCP server so it can pull the Google Ads data itself.

Combined with the lead data, the form data, the qualification data, all of that pulls in and Claude ha runs through a routine and goes, How do I need to look at this? What are the outliers? What are the anomalies? What's working? What's not working? And then makes a daily recommendation. Recommendation being the key word here.

Alane Boyd (11:08)
Yeah, so there are two parts that I wanted to bring up in this one. I want to talk about the heat map thing because it is so cool and gives you so much information on what people are doing when they're on your page. So I want to talk about that, Micah. And then there was one piece, and I I'm not gonna get it exactly right, but I'm hoping you know what I'm referring to, is in one of the keyword campaigns, we actually ended up splitting it into two different ones.

Because

of how people were converting and how we were tracking the keywords and the cost of those keywords.

Micah (11:41)
Yeah. So go ahead and ask your heat map questions first, 'cause like that's super

Alane Boyd (11:46)
Well, my my comment on

yeah, the heat map isn't so much a question more just helping people understand what we're referring to in it. And one of the things that we noticed is our landing page we thought was so beautiful and was so helpful. And there was a section where people thought that they could click on the things and it would like flip a card over or they could take an action. Yeah.

Micah (12:09)
yes. I've got a I've got a better

example than this one.

Alane Boyd (12:12)
Okay, okay.

Micah (12:14)
which is the the core to this whole system, which again, another episode because we won't have time in this one, is the idea that in order to do successful marketing you have to run a boatload of experiments. So in the database layer that we have in this system, it's experiment based. One of the early experiments that we ran.

was hey, I think we could do like an advertorial landing page, which is let's run some paid ads on specific topics that people are trying to solve, and we'll point them to a nice page that is kind of like a blog post, gives them a bunch of information, but ultimately to get a little further, then leads to a middle call to action and a bottom call to action. Because we had the heat map technology installed, Claude

Within one or two days, not weeks, of running this and going, God, why are we getting no leads? Claude was able to look at the heat map information and go, we have all these like rage clicks that are happening on these advertorial pages that we're paying for

Alane Boyd (13:20)
Mm, mm.

Micah (13:21)
traffic to go to, and we're getting no conversions on these pages. We're getting conversions on other pages, but not these. This is a contained experiment, right? And so we look and go.

All right, so Claude made the recommendation. Hey, look and figure out what the hell is happening with all these rage clicks. And the first thing we needed to do was go, what the hell's a rage click mean? It ultimately means, hey, somebody's trying to click on the page a bunch of times rapidly because it's not doing what they want to do. What we uncovered within an hour of just like, hey, let's analyze and see what's going on. Not just with Claude, we said, Claude, thank you very much. And then we went to

were using Microsoft Clarity, which is the heat mapping tool that we installed on our landing pages, and we were able to look at the reports in Microsoft Clarity and say, Hold on, these aren't necessarily rage clicks. These are people clicking on the text and dragging and you know, letting it go.

And

Alane Boyd (14:21)
Mm.

Micah (14:22)
so what we realized was we provided too much information on these

Alane Boyd (14:24)
That's right. Yes.

Micah (14:26)
pages that people were

Copying and pasting. So

Alane Boyd (14:30)
Mm-hmm.

Micah (14:31)
they were searching for a you know a problem that they had. They'd find our page. We'd pay for that click. And then they'd

Alane Boyd (14:37)
Yeah.

Micah (14:37)
go to our page, get the answer, copy and paste the answer, and move on with their day. And they never needed to click on the call to action because

Alane Boyd (14:45)
Yeah, yeah.

Micah (14:46)
they got their question answered.

Alane Boyd (14:49)
That's right, Micah. I forgot about that piece. I'm glad you remembered. We were being too valuable right there at the very start. So, but that is the that

Micah (14:55)
But it we

Alane Boyd (14:57)
is the the key to this, that we didn't just go

Micah (14:59)
We answered it with data.

Alane Boyd (15:01)
right. And having that heat map was so helpful for us to know and then make those tweaks.

So the heat mapping and the data was one piece of it. And then we also, while we're looking at the data, was the second part that I wanted to talk about, was when we split the campaigns.

Micah (15:17)
Yes. So we made some assumptions early on. And I think this happens to a lot of companies and a lot of businesses that are trying to figure out a solution to something, right? You have to just decide on something and move forward. So we decided on how we're going to qualify our leads. And we had a fully qualified, partially qualified, needs review and unqualified, which

Like we all talked about it. We go, that makes a lot of sense. This is what we, you know, we see all of our clients, we see what how people are engaging with us. This is all logical. But then we got about 300 plus leads, and that data allowed us to with that again, because we save this to a database layer, we can go to Claude and go, you know, something's not seeming right with our ad campaign. And

We were able to whittle it down in two ways. One, we could say by keyword, which one is generating more qualified versus unqualified leads? But the issue is we had fully qualified, partially qualified needs review. That system ended up being way too complicated. And fully qualified ended up being not what we deemed as fully qualified wasn't actually.

When we started getting the data, what we thought would be fully qualified based on historical data wasn't actually the case when we put it out into the real world on this campaign. So we did two things. We simplified it. We went, is it qualified or unqualified? One of the things that was getting qualified as fully qualified were individuals who wanted to upskill themselves or job seekers or students who just graduated.

It was more like individuals and in our mind we're so focused on businesses that we didn't even register that there's individuals in the world out there who might want to learn this stuff, which is ridiculous when we say

Alane Boyd (17:17)
Yeah.

Micah (17:18)
it. But

Alane Boyd (17:19)
We're in our silos too, people.

Micah (17:21)
Hey, but we got the data. We had Claude, because we had that data in the database and all the other data sources that Claude had it could cross-reference all of that and we could immediately well, we're talking like five or ten minutes of work of data analysis

Alane Boyd (17:36)
Mm-hmm.

Micah (17:36)
to go, that's what's happening. So we now split it up and we have qualified and unqualified. And

And then that helped us adjust and bifurcate the campaigns and tweak the keywords and say, let's spend less on this keyword or let's spend more on this keyword. Again, experiments. So each time we

Alane Boyd (17:58)
Mm-hmm.

Micah (17:59)
run these things, we log these experiments. Claude helps us log the experiments with the data. We then determine: did this experiment pass or fail? Or do we need to pause it? Claude helps us manage all of that. If we need to adjust the campaign, Claude helps us.

rewrite the CSV files, we upload it into the desktop app and push it to our ads campaign. And it's doing all that heavy lifting. But again, we're making the decisions on all of that. Claude and AI is not making judgment calls.

Alane Boyd (18:30)
Yeah, we we've got a lot of different tests happening too. And, you know, when we see, hey, something is kind of working for this lead set, let's move this over here and then have a different second landing page that they see if it's not a qualified lead. So we have a lot of pieces happening in the middle here that goes, hey, these things are kind of not working or kind of working, but as people and humans, like how can we strategize to make the

Really

because we want to get in front of those qualified leads. That's where we want to spend our time. how

Micah (19:03)
Mm-hmm.

Alane Boyd (19:03)
do we maximize those and help them instead of maybe these unqualified leads? And, you know, we can still give them a route that's helpful for them and some of those free resources that we have available in our community and things like that.

Micah (19:17)
Yeah, which what you're talking about

is like one of the latest experiments that's actively going on right now, Alane,

Alane Boyd (19:22)
Yeah.

Micah (19:23)
which is when a lead comes in, the second it comes in, it's being qualified. And before we show the post form submission page, we are

Alane Boyd (19:33)
Mm-hmm.

Micah (19:34)
routing them based on you know their qualification to

Alane Boyd (19:40)
Mm.

Micah (19:41)
personalized pages based on what

what information they've given us. That personalization stretches not just for the offers, but

Alane Boyd (19:50)
Mm-hmm.

Micah (19:51)
what the headlines and the subtext and some of the content on the page actually is, which is generating on the fly. Not not just

Alane Boyd (19:59)
Mm-hmm.

Micah (20:00)
boilerplate stuff, but actually taking what they're saying and generating a personalized page for them.

Alane Boyd (20:08)
Mm-hmm. And the the last piece of this that I wanted to kind of wrap it up with is well, how am I seeing all this information? Because like I said, you and the team built

Micah (20:17)
Mm-hmm, mm-hmm.

Alane Boyd (20:18)
it and I wasn't a part of it. Well, how do I get the information without being part of the build? So we have two different things that happen. One is we set up a private Slack channel for the leads coming in and where the analysis goes. So that daily review and everything just goes in there and then.

leadership team that needs or wants to be involved with that can see that when it comes through. So that's the way that I see that like end result. The other thing is that We have a scorecard for

our leadership team, but it's a dashboard. And so

You know, I'm getting that kind of like granular daily review in Slack, but like overall, where are we? And in our dashboard. So we've got that other view in Claude Cowork as a shared dashboard for our leadership team. So I'm also able to see the bigger picture there.

Micah (21:06)
Yeah, I think I'm really glad you pointed this piece out, Alane, because if we frame it slightly differently, together as a leadership team, we said here's the goal we're trying to achieve. From a technical perspective, you're playing the role of stakeholder in this case, right?

Alane Boyd (21:24)
Mm-hmm.

Micah (21:25)
I'm playing the role of lead developer, so to speak, and we can have

the technical side taken care of without the other stakeholders necessarily needing to be involved. But

we

Alane Boyd (21:37)
Mm-hmm.

Micah (21:38)
don't have to have like all these crazy meetings because you're getting the daily updates in Slack. And once a week in our normal leadership goals call, we have a dashboard scorecard that again, because we're tracking the data layer, surfaces everything that we need. There's no manual like.

Zero manual work required to understand where everything is at based on the objective we set out to achieve.

Alane Boyd (22:08)
Mm-hmm. I love both ways that we have it too, because on the daily review piece, if we want to have a quick brainstorm based on the analysis that's given to us from the ads, then we can and we can say, Hey, what if we adjust this tomorrow? What if we do this? Then that's a great like little in-housed area. And then for the scorecard in our dashboard, we can just pull that, like we can look at that anytime.

Cause it's right there in our on our pinned dashboards in Cowork. But for our leadership meeting, nobody's having to go and pull that data. It's already ready to go.

Micah (22:42)
Yeah, we don't pay anybody to pull that data. We don't use our time to pull that data. We built it with Claude and Cowork, which maybe took 30 minutes.

Alane Boyd (22:52)
Ha ha.

Micah (22:53)
And that was it. And I mean, honestly, for anybody listening to this episode, we covered a ton in this. I would say key takeaways are it's leveraging Claude to do the heavy lifting or AI to do the heavy lifting with human judgment.

It's when you're building systems like this, start small and iterate in a piece at a time. It's save the data in a database that you need so that you can actually build these dashboards quickly. The reason that we could build all this stuff very quickly is because we structured it well. And if you need help or want ideas on, you know, somebody to bounce ideas off of on your own systems, reach out to us. That's what we do for clients.

Alane Boyd (23:37)
Mm-hmm. And I'm gonna, I'm gonna do you one better, Micah, because we have deep dive workshops that we have coming up that we are teaching on how exactly we set all of this up. So it part one on August 6th, where we're talking about how we set up this Google Ads with Claude entire system.

And then part two, because there's a lot here. We can't all be done in one deep dive. The second one is on August 12th. So if you're interested in getting this set up, doing everything that we just walked through, come and join us on August 6th and August 12th.