-
My experience using Claude to actually manage Google Ads
Posted by on April 21, 2026 at 6:10 amHeads up: this is an opinion post, not a tutorial. Plenty of setup guides already exist. The content is rephrased by AI.
Context: I run a multi-location service business. 8K/month budget. Former software engineer, so I tend to build my own internal tools instead of buying. I've been using Claude to manage my Google Ads for a while, and by manage I mean read and write. Pausing keywords, adjusting bids, restructuring campaigns, writing new RSAs. Not just summarizing last week's performance.
I think this is paradigm shift in how ads are going to be managed. It is a bit early right now, but the as models get better, I just can't see this reversing. But it's worth being precise about who it's actually for, because most takes I see online are either "AI replaces all PPC tomorrow" or "AI is useless for ads," and neither is right.
Who this won't help
If you've never run a campaign and don't understand how paid search works at a structural level, don't hand Claude the keys. Not because it isn't capable, it is, but because it doesn't have your business context. It doesn't know your margins, your seasonality, which leads actually close, what a defensible CPA looks like for you. Without that, you can't evaluate what it gives you. You'll get plausible-sounding recommendations and no way to validate them. You're better off hiring someone competent.
It also doesn't replace strong agencies. Senior media buyers do a lot of work no tool touches: strategy, creative direction, managing Google rep relationships, fighting policy disputes, knowing when to ignore the platform's recommendations. That's not going away at least for now.
Who it's genuinely useful for
Two groups, in my experience.
Business owners who already run their own ads. If you connect your CRM, content management system, google search console, GA4, and Google Ads so Claude Opus can see all of it together, you will get some pretty amazing result because the top models can synchronize and analyze all these data and produce very professional analysis. For example, flagging that a search term is converting on the ads side but those leads never close in your CRM, which means you're paying for the wrong intent. That kind of cross-system analysis requires some expertise and technical ability. It's now within reach for an operator who knows what to ask.
A concrete example from my own account: my business offers several distinct services, but my original campaign had all the keywords lumped into one campaign with no real alignment between keyword intent, ad copy, and landing page. Quality scores were predictably mediocre, which meant I was paying more per click than I needed to. Claude restructured the account properly, separated campaigns and ad groups by intent, rewrote ad copy to match each group, and even built out the dedicated landing pages so the whole funnel was actually coherent. That's not a small task that I want to prioritize especially when I am not 100% certain of the return on my time. But with Claude, the marginal cost of making these changes are 0, so I am happy to have it do it all.
You see how this is shifting the economics – without Claude, at my budget, no agency or freelancer is going to do deep work on every little thing and even help me change my content and my website. The economics don't support it. That's actually where AI changes things most: it makes that depth of analysis viable at budgets where human help never made sense. So if anything, smaller advertisers benefit more, not less, as long as you know enough to direct it.
Agencies. This is the case I think is most transformative, and I'm not sure how many agencies have really sat with the implications yet.
If you run an agency, you already have a playbook. How you audit a new account, how you decide whether to restructure or optimize in place, your weekly reporting format, your QBR structure. The hard part isn't knowing what to do, it's executing that playbook consistently across a roster of clients.
That's the kind of work Claude Code is well suited for. Encode the playbook as a skill or plugin, and a single operator on a Max plan can produce genuinely customized weekly deliverables for every client. Not template output with the company name swapped, actual analysis grounded in each account's data, with recommendations that reference real numbers and account history.
The downstream effect on agency economics is what makes this interesting. Smaller accounts become profitable to serve properly because the marginal cost of a thorough review drops a lot. Headcount scales more slowly relative to client count. And the quality floor goes up, because every client gets the playbook applied consistently rather than depending on whichever AM happens to be sharp that week.
Curious whether others here are doing this, and what's working or not working. Happy to go deeper in the comments.
-
1 Reply
-
interesting approach!
I have started moving this way but from SEO side – building an internal tool to track keyphrases, fix pages in semi-auto mode, address page load speed issues etc. Now focused on internal links suggestions and backlinks analysis.
I am wondering if apart from playbook are you using some special preset skills or instructions for ads management, or do you just trust “bare” Opus model, especially for ads analytics, things like when to start/stop keyword etc?
-
Yeah this matches what I’ve seen pretty closely. What matters is whether the person using it already understands paid search at a structural level. Search intent, account structure, conversion tracking, margins, lead quality, all of that. If they do, Claude becomes leverage. If they don’t, it just gives them very confident sounding bad ideas**😅**
Looking at Google Ads alone only gets you so far. Once the model can see CRM outcomes, landing pages, GA4, Search Console, call data, maybe even sales notes, it starts thinking about the juicy stuff. Not “this keyword converted.” More like “these leads never close” or “this ad group looks efficient in-platform but the intent is wrong once you follow it through.”
I also agree on the agency angle. If you already have a real playbook, the economics change quickly. Smaller accounts that were never worth deep manual work suddenly can get proper attention daily through automated actions. That does not replace senior buyers. It just changes how much one good operator can handle without quality slipping.
Only thing I’d add though is that the write side needs guardrails. Read-only analysis is the easy part. Letting it push changes live is where you find out fast whether your context, tracking, and operating rules are actually solid.Feels a year early for the average advertiser, but I don’t see this going backwards either.
-
Great post. Yesterday I build a similar thing with codex and Chatgpt plus and Google Cloud console.
I already had the api’s from GSC, GTM and GA4 working. Only the Google Ads API access I had to request basic access. Did that take a long time?
For now I used only read only from the enabled api’s. Google Ads seems to be read and write, but I instructed it clearly with skills and agents.md to only read for now.
Are you comfortable with Claude Opus touching your daily budgets?
-
I have used a lot of AI tools over the course of the past 2 years and I can say that there is no tool that helps the business scale effectively. In fact, most of them suffer.
That said, claude skills are great at time management but they still miss the mark in some critical tasks such as negative keyword lists.
Hope it helps
: ) -
Really appreciate how nuanced this take is, most people are either overhyping or dismissing AI in ads entirely.
I’m running ads across Google Ads, Meta, and TikTok for multiple clients (agency side), and what you said about execution vs. strategy really hits. The playbook already exists, the bottleneck is consistent, high-quality execution across accounts.
The part that stood out most to me is the cross-system analysis. That’s something even good media buyers struggle to do consistently unless everything is tightly integrated (CRM, attribution, lead quality, etc.). If Claude can actually bridge that gap in a reliable way, that’s a big shift.
Curious about your setup though:
* Are you connecting Claude directly to tools like Google Analytics 4, Google Search Console, and your CRM via API, or are you feeding it exports / structured data manually?
* How “real-time” is your workflow? Is Claude actively making changes, or are you reviewing and pushing live yourself?
* What niche are you operating in, and what campaign objectives are you mostly optimizing for (leads, calls, bookings, etc.)?
* Have you tested how it handles budget scaling decisions or more sensitive changes like bid strategy shifts?I’m seriously considering building something similar internally for our agency, especially for audits, restructuring, and weekly reporting.
Would love to hear more about how far you’ve pushed it.
-
Claude should be used as an analyst, not as a buyer role. There should always be a human at the helm approving any changes to an account.
-
This is 100%. if you dont know your margins claude will just help you go broke faster with plausible sounding recommendations. Business context is the only guardrail that matters.
-
the landing page rebuild is the part people skip over, matching ad group intent to a dedicated page moved my quality scores more than any bid change and having claude do the copy at zero marginal cost is what made that actually happen at my budget
-
How are you finding the quality of the numbers here?
My experience so far is that if you give LLMs access to raw data and ask them to produce analysis, they quickly begin to hallucinate nonsense as soon as you move beyond reporting the basic figures – they don’t math well. We pay for 3rd party tools that have added AI analysis and it quickly devolves into rubbish when you ask it for anything beyond the absolute basics.
Because of this, I built my own system that pulls and processes raw performance data to produce a whole suite of derived stats and metrics before sending it all to the LLM for analysis. So far, the results of this have been pretty bullet-proof, but I still review them and make optimisation decisions personally.
So, I guess my question is: what guardrails do you have to ensure that the AI isn’t basing optimisation decisions on hallucinated data, especially when asking it to make decisions based on information from multiple platforms?
Log in to reply.