AI ads infrastructure

Google Ads AI Copilot: Suggests, Drafts, You Approve

An AI copilot for Google Ads sits next to you — drafting changes, surfacing waste, executing on approval. Here's how it differs from Smart Bidding.

NotFair Team|

An AI copilot for Google Ads is exactly what it sounds like — an AI sitting next to you in your account, suggesting changes, drafting bid updates, and executing on your approval. Same idea as GitHub Copilot for code: the human stays in the driver's seat, the AI handles the tedious work, every change is reviewable. It's a different category from Google's Smart Bidding (which is autopilot, not copilot) and from agency software (which is dashboards, not collaboration).

The product most marketers use today is Claude or ChatGPT plus a Google Ads MCP connector. NotFair is the hosted connector — it exposes about 40 tools backed by the Google Ads API and gates every destructive write behind your approval. The result is a copilot you can talk to: ask it questions, get suggestions, approve or reject each change.

Copilot vs autopilot: the important distinction

Smart Bidding is autopilot. You set a target, Google optimizes bids inside its black box, and you don't see the per-decision rationale. It's powerful but opaque. A copilot is the opposite: every action is initiated by a human prompt, the rationale is visible ("I'm pausing these because they had $1,200 in spend and zero conversions"), and you can approve, modify, or reject individually.

  • Smart BiddingOptimization decisions made inside Google's models. You set objectives. You see outcomes. Limited insight into per-bid rationale.
  • AI copilotDiagnostic and editing decisions made through conversation with you. You see every proposed change before it ships. Full audit trail of what was changed and why.
  • Agency dashboardsReporting layers on top of Google Ads. Pretty charts, no agency in the chart.
  • Browser-automation agentsAI driving the Ads UI. Slow, brittle, and skips the API safety net. Avoid for production work.

What a Google Ads copilot actually does

  • ReadsAudits, search-term reviews, change-history queries, quality-score breakdowns, performance dashboards. Runs autonomously since reads can't break anything.
  • DraftsWrites proposed changes (paused keywords, new negatives, bid updates, RSA copy variants) but does not execute until you approve.
  • SuggestsSurfaces "three quick wins" or "the one bid change that would most improve ROAS" based on account state.
  • ExecutesOn approval, calls the Google Ads API directly. No clicking through the UI. No CSV uploads.
  • ReversesMost operations expose an undoChange tool. If a change was a mistake, the rollback is one prompt.

What it feels like to work with a copilot

You: Anything I should know about this morning?

Claude: Three things: 1. Yesterday's spend on Search - US was $214 above pacing because impression share spiked overnight. Causes: a competitor paused a Brand campaign and a new search term "crm comparison 2026" started matching broad. CTR is healthy (4.1%), CPA is fine ($38). No action needed unless you want to cap budget. 2. The Black Friday RSA you A/B tested last week is statistically significant — Variant B has 23% better CTR. Want me to pause Variant A and shift impressions? 3. Three keywords in Search - APAC dropped to QS 2 in the last 24 hours. Likely landing-page experience flagging on a recent product page change.

You: Pause Variant A and look at the APAC quality score issue.

Why "still in control" matters here

Many marketers we talk to are nervous about handing AI the keys. The valid concern: an AI that misreads context can torch a budget overnight. The copilot pattern solves this directly. Read access can run autonomously — it's safe. Write access never runs without you. Every paused campaign, every bid update, every negative keyword is a diff you saw and approved. The AI is a force multiplier, not a replacement.

The four tiers of trust to graduate through

  • Tier 1 (week 1)Read-only audits and diagnostics. Get comfortable with the model's reasoning before granting any write.
  • Tier 2 (week 2-3)Approve writes one at a time. Review every diff. Use this period to calibrate trust on small-stakes changes (negatives, RSA edits).
  • Tier 3 (month 2)Batch approvals on low-risk operations (negatives, paused-keyword adds). Still gate budget and bid-strategy changes individually.
  • Tier 4 (month 3+)Recurring autonomous tasks for read-only reports. "Audit my account every Monday morning and email the summary" runs without you.

Where a copilot beats the alternatives

  • vs hiring an agencyA copilot runs the diagnostics an agency would charge $1,500 for in 60 seconds, anytime you want, with the same quality.
  • vs hiring a junior media buyerA copilot handles the 80% of work that's mechanical (search-term cleanup, negative adds, bid maintenance), freeing humans for strategy.
  • vs Google Ads EditorEditor is a fat client for bulk edits. A copilot is a conversational layer on top of every API operation, including diagnostics Editor doesn't support.
  • vs scriptsGoogle Ads Scripts require code. A copilot writes the GAQL for you and explains the result. Same power, no engineering effort.

Setting up your copilot

  • Sign in at notfair.co/connect with the Google identity that has Ads access.
  • Pick the customer IDs the copilot can seesingle account or every account in a manager (MCC).
  • Install the connector in your AI client of choiceClaude Desktop, Claude Web, ChatGPT (Codex connectors), Cursor or Windsurf (stdio config).
  • First prompt: "audit my account." From there, copilot work is just conversation.

FAQ

Try MCP with Google Ads

Connect your Google Ads account to NotFair in 30 seconds and start querying campaigns from Claude.

Connect Google Ads

FAQ

Common questions about Model Context Protocol.

How is a Google Ads AI copilot different from Smart Bidding?

Smart Bidding is autopilot — Google optimizes bids inside its model with limited per-decision visibility. A copilot is collaborative — you initiate every action, see every proposed change, and approve before execution. Different categories, both useful.

Will the copilot make changes without my approval?

No. Reads run autonomously. Every destructive write — pause, bid change, delete — passes through a server-side approval gate. You see the diff and confirm before anything ships.

Can I roll back a change the copilot made?

Yes. Most operations expose an undoChange tool, and the change history is queryable through the same conversation. "Roll back the negatives I added yesterday" is one prompt.

Does the copilot replace my media buyer?

No. It handles the mechanical 80% — search-term cleanup, negative adds, audits, bid hygiene — so humans can focus on strategy, creative, and account growth. Most teams add a copilot, they don't subtract a person.

Which AI clients can I use as my copilot?

Claude Desktop, Claude Web, Claude Code, ChatGPT (via Codex connectors), Cursor, and Windsurf all support MCP and work with NotFair. Pick whichever your team already uses.