This Openclaw agent manages my Google Ads

Last Tuesday I pointed this Openclaw agent at a Google ads account I was asked to audit. 4,300 search terms. $23,000 in spend over 90 days.

First thing it flagged wasn't waste like I thought. It was a rotten CPA.

The account showed $179 cost per acquisition. Respectable. The kind of number you'd look at in a dashboard and think, okay, we're fine. But the agent broke it down by intent class, and the real story was ugly:

$4,200 a month going to "what is crm" and "crm jobs near me" and "free crm template download." 847 queries. Zero conversions. Just... gone.

A few weeks ago I released my Openclaw agent that runs your Meta ads. And I kept getting one question… “do you have one of these for Google ads?”

Now I do.

I built an Openclaw agent that runs your Google ads. Fifteen skills. Full intent mapping. Daily operator reviews. And I'm giving the whole thing away free (scroll to the bottom for the link).

Here's how it works.

Every account I tested had a version of the same problems. Here's what the agent finds:

When buyer queries, comparison queries, and educational queries all live in the same campaign, Google reports one blended CPA.

That number is meaningless.

The agent breaks it apart by intent class.

Every account I tested had a fake CPA. Every single one.

Search partner traffic eating 12% of spend with 0% of conversions.

A 72-hour conversion lag that makes yesterday's data useless.

Double-counted form submissions.

None of this shows up in the Google Ads dashboard.

The agent goes looking every morning.

This is the big one. Someone searches "clio vs hubspot crm." That's comparison intent. They need different copy, a different landing page, a different CPA target than someone searching "crm demo for law firms."

But in most accounts? Same campaign. Same ad group. Same ads. Same landing page.

Google's algo sees both and optimizes for the average. Both experiences are wrong. Both waste money.

Most Google Ads problems aren't bid problems. They're intent-mixing problems. The agent reads 4,000 search terms and classifies every one by intent in 40 minutes.

Once you see the intent map, you see the landing page problem immediately.

Comparison shoppers hitting your homepage. Buyers hitting a blog post. Branded searches going to a generic page.

The search term told you exactly what the person wanted. The account structure ignored it.

This is the part that makes it different from every audit tool I've used.

The agent doesn't just tell you what's wrong.

It drafts specific fixes (negatives with match type and scope, structure changes, budget moves) and stages them for your review.

You approve. It applies. Dry run first. Full audit trail. Instant undo.

Telegram message from the agent:

I read that in two minutes. Approve the negatives. Greenlight the comparison split. Move on.

Total time managing Google Ads that day: under 5 minutes.

The agent didn't give just recap the numbers. It gave me decisions to act on.

Every time the agent reviews your search terms, it builds something I call the Intent Map. A living model of what your Google Ads account is actually buying.

The map lives in your workspace. It persists across sessions.

Most PPC tools give you a snapshot. This gives you memory.

"What is [industry term]" is almost always waste. "[Role] jobs near me" is always waste. "Free [product category]" converts at near-zero everywhere.

Over time, the agent builds the same pattern library a senior strategist carries in their head. The difference is it doesn't forget.

Fifteen skills. Three layers.

Read layer pulls live account data (search terms, keywords, conversions, budgets) through Google's official MCP server. Or you paste a CSV export. Same analytical engine either way.

Draft layer turns findings into concrete proposals. Not just a "your CPA is high."

These are specific negatives with match type and scope.

Structure changes with before/after projections.

RSA directions pulled from the language real buyers actually use. Every proposal staged in workspace/ads/drafts/ for your review.

Apply layer executes what you approve. Dry run shows you exactly what will change. You confirm. It applies.

Full audit trail logged. Instant undo with /google-ads undo if anything looks wrong.

The difference between this and a thermometer: a thermometer tells you the temperature. This tells you why you have a fever and hands you the medicine.

Google wants you to trust their automation. Broad match. Smart Bidding. Performance Max. "Let the algorithm do its thing."

Smart Bidding is genuinely good at auction-time decisions. But Google's AI optimizes for Google.

They push broad match because it expands inventory. They push Performance Max because it lets Google decide where your money goes. They changed the search terms report to hide "low volume" queries you're paying for but can't see. They build recommendations that auto-apply if you don't manually opt out. Every update in the last three years has reduced advertiser visibility and control while asking for more trust and more budget.

When Google says "let our AI manage your ads," ask: manage them toward what outcome?

The winning move: use both. Let Google handle the auction. Use your own agent to manage the strategy layer on top. Read the search terms. Build the intent map. Cut the waste. Shape the structure.

You keep the judgment. They run the auction.

Step 1: Get the kit

Fifteen skills. Reference docs. Workspace templates. Apply layer with audit trail. Everything.

Step 2: Connect your account

Two options. Connected mode hooks up Google's MCP server and the agent pulls live data automatically. Export mode: paste a CSV from Google Ads. Same analytical engine, same quality. Connected is better for daily ops. Export is fine for a first audit.

The agent discovers your accounts, picks one, writes the workspace config. If you're using export mode, skip this and go straight to Step 3.

Step 3: Run your first Query Interview

Drop in your search terms report (or let the MCP pull it live). The agent reads every query, classifies intent, finds waste, and builds your first Intent Map. This is the moment. 4,000 rows of search term chaos turned into a clear intent model with specific cut/isolate/scale recommendations.

Step 4: Run the daily operator

The morning review. Set it on a cron schedule. The brief lands in Telegram or wherever you want it.

Step 5: Review, approve, apply

The agent drafts. You review. Approve what makes sense. The apply layer dry-runs the change, you confirm, it executes. Full audit trail. /google-ads undo if you change your mind.

Start with negatives. It’s lowest risk, highest immediate impact. Graduate to structure changes when you trust the reasoning.

The difference is the difference between a thermometer and a doctor. One tells you the number. The other tells you what it means and what to do about it.

Let me be direct: if you know your business (your offer, your margins, what a good customer looks like) this replaces most of what you're paying a PPC agency or consultant to do.

Reading search terms. Classifying intent. Drafting negatives. Flagging waste. Monitoring tracking health. Recommending structure changes. Writing RSA directions from real buyer language. Applying changes with audit trails.

That's 90% of the billable hours in a $3-5K/mo agency retainer. It's execution work billed at strategy rates.

This agent does all of it. Every morning. With memory.

Where it doesn't replace a human: knowing whether to run Google Ads at all. Knowing what your landing page should say. Knowing when a $300 CPA on comparison queries is a problem or an investment in a new market. That's strategy. That's the 10% that's genuinely hard.

But if you're paying someone $200/hr to scroll your search terms report? That's not strategy. That's maintenance. And you don't need a human for maintenance anymore.

Not for you if you want full auto-pilot or expect bid optimization. This is intent interpretation. A different and more fundamental problem.

The full system is open source:

15 OpenClaw skills. Read/Draft/Apply pipeline. The Query Interview framework. Intent Map that persists and compounds. Workspace memory across 12 persistent files.

Apply layer with dry run, audit trail, and instant undo. Connected mode via MCP or CSV export. Setup guide from install to first automated review.

No catch. No upsell. MIT license. Fork it.

If you'd rather have someone run this against your account and show you what your search terms are actually saying, that's what we do at Emerald. Reply and we'll talk.

Let’s go big,
Matt

P.S. That new client with the $4,200 in monthly waste? Cut in the first week. Their real buyer CPA: $112. Not the blended $179 the dashboard showed. No bid changes. No new keywords. Just stopped mixing unlike things together and let the numbers tell the truth.

P.P.S. If you missed the last three giveaways (the Meta Ads agent, the First 1000 Customers kit, and the SEO Kit) they're still free on GitHub.

🎁 Get rewarded for sharing! My team grew NEW accounts to over 50 million views in just a few months. We made an AI viral hook generator so you can follow the same hook frameworks that we do.

Invite your friends to join, and when they sign up, you’ll unlock our AI Viral Hook Generator—the ultimate tool for creating scroll-stopping content.

{{rp_personalized_text}}

Copy and paste this link: {{rp_refer_url}}