Talking to Google Ads With AI: What It's Like

Imagine opening a chat window, typing "why did my CPA jump 40% last week?" and getting a real answer in 30 seconds — not a generic article, not a dashboard URL, but the actual diagnosis from your actual account, with the specific campaigns and search terms that broke. That's the experience people are looking for when they search for "talk to Google Ads with AI." The good news: it exists, it works today, and you can set it up in five minutes.

Here's what a real five-turn conversation looks like with Claude connected to a Google Ads account. We'll explain the technology underneath after.

What just happened under the hood

That conversation involved roughly six tool calls behind the scenes. Claude pulled campaign-level performance for the trailing two weeks, fanned out parallel queries across search terms and change history, ran a Python-style filter on the result, drafted a write operation, surfaced an approval prompt, and finally executed the negative-keyword additions through the Google Ads API. The whole thing took 45 seconds of wall clock time.

The piece that makes this possible is called MCP — Model Context Protocol. Anthropic published it in late 2024 as an open standard for connecting AI models to external tools and data. NotFair is a hosted MCP server at notfair.co/api/mcp that exposes about 40 tools backed by the Google Ads API. When Claude needs to read your account or apply a change, it calls those tools through the protocol. You never see the API plumbing — you just type and read replies.

Is this a real product or a demo?

It's a real product. Real accounts, real spend, real changes flowing through Google's API the same way an agency dashboard would push them. The conversation above is representative — we run it on our own accounts daily. The only thing that varies is how much context the AI has about your business goals, which improves the more you talk to it.

Where the magic actually shows up

• Diagnosis — "Why did conversions drop?" used to be a 90-minute spreadsheet exercise. Now it's a 45-second answer with the specific change events, search terms, and quality score moves that explain the drop.
• Bulk operations — Adding 50 negatives across 12 campaigns is a sentence, not a CSV upload. Pausing every keyword under $0.50 quality score with zero conversions in the last 30 days is one prompt.
• Hypothesis testing — "What would happen if I cut budget on Search - EU by 30%?" The AI can model historical impression share, project lost conversions, and tell you whether the trade is worth it before you make the change.
• Audit on demand — "Audit my account" returns a structured report with waste, quality score gaps, search-term hygiene issues, and ad-copy underperformers in under a minute.

The parts that aren't magic

Conversations with AI feel magical, but the AI is still operating on the raw Google Ads API. If a metric isn't in the API — say, attribution model nuances or audience overlap from third-party tools — the AI can't see it either. Same with strategic decisions. The AI will recommend reallocating budget from a losing campaign to a winning one, but the call to actually shift dollars stays with you. That's a feature, not a bug.

Approval gates: the part that keeps you safe

Every destructive write — pausing campaigns, bulk-updating bids, deleting ads — passes through a server-side approval gate. The AI drafts the change, you review the diff, you approve. There is no "the AI went rogue and burned my budget overnight" failure mode because the AI cannot execute writes without your explicit confirmation. Read operations (audits, diagnostics, dashboards) run autonomously since they can't break anything.

How to start a conversation with your ad account

• Sign in at notfair.co/connect — OAuth scoped to Google Ads only, takes 30 seconds.
• Pick which customer IDs you want the AI to access. You can select one account or every account in a manager (MCC).
• Install the NotFair connector in Claude Desktop, Claude Web, or ChatGPT (Codex connectors). Cursor and Windsurf work too via stdio config.
• Open a chat and ask "audit my account." The first audit takes about a minute. From there it's just conversation.

Try the openings that everyone uses on day one

• Find wasted spend — "Find every search term over $50 in spend with zero conversions in the last 30 days."
• Diagnose a regression — "Why did CTR on Search - Brand drop this week?"
• Quick wins — "Give me the three changes that would most improve ROAS in this account."
• Daily standup — "Summarize what changed in this account yesterday and the impact."

FAQ