GoogleAds Analyst Skill

Problem

Asking an LLM “how is my Google Ads account doing?” gets you a plausible-looking report with no discipline: invented metrics, no distinction between what you changed and what the market did, and no safe path from “here’s the problem” to “here’s the fix.” Analysts need a repeatable methodology, not vibes.

Goal: turn Claude into a Google Ads performance analyst that follows a real investigative workflow — and can act on its recommendations without ever surprising the account owner.

Constraints

1. LLMs hallucinate GAQL field names. metrics.video_views vs metrics.video_trueview_views looks right and fails at runtime.
2. Attribution is the hard analytical problem. A conversion drop after a bid change is a different problem than the same drop with no account activity.
3. Mutations are irreversible-ish. Pausing a campaign is recoverable; removing one is not. A wrong write is worse than no write.
4. Token budget. A skill that front-loads its entire methodology into context is slow and degrades the model’s attention.

Decisions

• Validate queries before execution. Every GAQL query must pass mcc-gaql --validate against API metadata before execution. This catches invented field names before they reach the account.
• Load guidance when needed. The skill loads analysis guidance as each phase needs it, keeping the initial context small.
• Compare anomalies with account changes. Timing, symptom match, magnitude, and affected scope help distinguish the effects of account changes from external factors.
• Require approval before writes. The skill inspects the current state, previews the change, asks the user to approve the exact action, applies it, and verifies the result. Irreversible removals get additional safeguards.
• Treat Grants accounts differently. Google Ads Grants accounts have different spending and eligibility constraints. Dedicated thresholds keep the analyst from flagging normal Grants behavior as a crisis.

Artifacts

• Repo: mhuang74/googleads-analyst-skill
• Companion CLIs: mcc-gaql-rs provides query, validation, generation, and mutation tools

Outcome

In monthly use since 2025: generates the monthly performance report for three Google Ads accounts.