google-ads is the operating contract for managing a Google Ads account with an AI agent. It reads live evidence through the connected MCP, reconciles metric definitions, separates fact from inference and states a decision rule behind each material recommendation.
Changes to budgets, bids, statuses, targeting or any spend-affecting setting need the user’s explicit yes to the exact change. After an approved change the agent verifies the result against live data and records it so a later impact review can decide whether to keep or revert it.
What You Get
- Performance diagnosis with account currency, dates and denominators shown
- Search-term triage into negatives, keyword candidates, routing issues, winners and watch items
- A weekly comparison of each campaign’s last 7 complete days against the prior 28
- Approval-gated changes with live read-back verification
- Handoffs to the audit, copy and landing page skills when the data points there
Common Use Cases
- Finding wasted spend and adding negative keywords
- Deciding whether a campaign should scale or hold
- Choosing or migrating a bid strategy
- Running a recurring daily check of keywords, pacing and tracking
What Makes It Unique
It treats judgment as the main tool: the agent chooses query shape and depth, and a library of reference files supplies calibration only when needed (decision quality, decision math, quality score, bid strategy, search-term triage, safe execution). The default posture is the smallest reversible action, and for small accounts it prefers a watch note to a mutation. Every material change is recorded so 3, 7 and 14 day reviews can decide keep, revert or iterate.
Under The Hood
The skill is a short wrapper whose SKILL.md tells the agent to read a longer WORKFLOW.md in the google-ads manage folder and follow it, stopping with a packaging error if it cannot. The workflow is prose that points to many reference and shared files in the repository and ships no scripts. It calls the connected NotFair MCP to run reads and queries against the Google Ads account over the network, and it writes spend-affecting changes only after explicit approval. It also reads and writes business context files in a data directory, and treats ad text and tool output as untrusted data.