ads-audit（广告审核）

Google Ads Audit

Diagnose account health and persist business context for downstream skills (/ads, /ads-copy, /ads-landing). Read-only — never mutates the account. The user runs /ads to execute fixes you recommend.

Setup

Follow ../shared/preamble.md — MCP detection, API key, account selection.

Filesystem contract (MUST persist)

These are the handoff to every other ads skill — write them even if the report itself is short. Otherwise /ads-copy and /ads-landing operate without business context and produce generic output.

business-context.json schema: business_name, industry, website, services[], locations[], target_audience, brand_voice{tone, words_to_use[], words_to_avoid[]}, differentiators[], competitors[], seasonality{peak_months[], slow_months[], seasonal_hooks[]}, keyword_landscape{high_intent_terms[], competitive_terms[], long_tail_opportunities[]}, social_proof[], offers_or_promotions[], landing_pages{}, notes, audit_date, account_id.

personas JSON schema: {account_id, saved_at, personas: [{name, demographics, primary_goal, pain_points[], search_terms[], decision_trigger, value}]}. See references/persona-discovery.md.

Policy freshness check (run first)

Read ../shared/policy-registry.json. For each entry where last_verified + stale_after_days < today:

• High-volatility → WebSearch the area for recent Google Ads changes; compare to assumption. If drift, banner the report and suggest registry update.
• Moderate-volatility → one-line "may warrant a check" note.
• Stable → skip silently.

Phase 1 — Pull the audit dataset

Use a single runScript call with ads.gaqlParallel to fan out the queries an audit needs. The server's adsagent://playbooks/audit-account resource has a battle-tested baseline; extend it with what this rubric needs that isn't already there.

A complete audit needs at minimum:

• Account / customer-level metrics (customer resource) — totalSpend, conversions, conv-value, clicks, impressions for the audit window.
• Campaign-level performance (campaign) — id, name, status, advertising_channel_type, bidding_strategy_type, network settings, search/top/abs-top impression share, budget-lost-IS, rank-lost-IS, status-rich metrics. Cap window at 90 days (impression-share data limit).
• Ad-group performance (ad_group) — to attach top-line metrics under each campaign.
• Keyword performance with QS (keyword_view) — text, match type, status, quality score (and components if needed), cost, clicks, conversions. Surfaces zombie keywords (0 impressions 30d), low-QS-with-spend, zero-converters.
• Search terms (search_term_view) — for waste detection (clicks > 10, conv = 0) and brand-leakage detection (brand terms triggered by non-brand campaigns).
• Negative keywords (campaign_criterion where type='KEYWORD' and negative=TRUE) — for negative-conflict detection and coverage.
• Conversion actions (conversion_action) — count, status, primary-vs-secondary, counting type. STOP-condition input.
• Ad assets (ad_group_ad) — RSA headlines/descriptions for asset-coverage scoring and topAds extraction.
• Geo targeting (campaign_criterion where type IN ('LOCATION','PROXIMITY')) — for multi-location structure scoring. radius_units: 0=meters, 1=km, 2=miles.
• Recent change events (change_event, max 30 days) — to flag recent edits that might explain regressions.

Compute aggregates in the script, return summarized JSON. Don't return all rows — rank, slice, summarize. The agent narrates the result, the script does the math.

getRecommendations is a useful cross-check for Google's own recommendation surface — call it as a separate tool after the runScript pass if you want to compare your findings to Google's.

If a critical query errors out (auth, schema), surface the error and stop — don't fall back to a degraded audit.

Skip scoring entirely if totalSpend == 0 or activeCampaigns == 0. Go straight to business context.

Phase 2 — Scope handling

If the user narrows the audit ("focus on grooming", "campaign X", "just check waste"):

• Match campaign names by case-insensitive substring. If no match, list available campaigns and ask.
• Filter the in-memory dataset before scoring — no extra API calls.
• Account-level dimensions (conversion tracking) stay account-wide. Note "Scoped to: X" in the report.
• Skip Phase 4 (business context refresh) on scoped audits if business-context.json is fresh.

Phase 3 — Score

Score each of the 7 dimensions 0–5 using references/account-health-scoring.md. Overall = round(sum × 100 / 35).

Scope-aware: campaign-level dimensions reflect in-scope data; account-level dimensions (conversion tracking) score account-wide with a note on scope impact.

Encoded heuristics — apply these, they aren't obvious

• Weighted QS by spend, not by keyword count. A QS-3 keyword burning $2,000/mo matters infinitely more than ten QS-3 keywords burning $5/mo combined.
• Brand-leakage premium. When brand campaigns are paused or starved, brand traffic leaks to non-brand at 5–10× higher CPA. Always check whether brand terms appear in non-brand campaigns' search-term reports.
• Waste formula. Keyword waste = clicks > 10 AND conversions = 0 AND cost > 0. Search-term waste = same, applied to actual queries. Sum them — that's the recoverable spend.
• Display + Search mixed in one campaign is structurally broken — Display dilutes Search metrics and burns budget on unintended placements. Flag any campaign where network_settings.target_content_network = TRUE for a SEARCH channel campaign.
• Zombie keywords (0 impressions for 30+ days) clutter reporting and confuse Google's ML — recommend pause.
• Counting type matters. Lead-gen should use ONE per click; e-commerce should use EVERY. Wrong setting silently inflates or deflates conversions.
• STOP condition. If conversion tracking scores 0–1, recommend pausing spend until tracking is fixed. Everything downstream of measurement is unreliable.

Impression Share Interpretation Matrix

Phase 4 — Business context

Derive what you can from the data already pulled:

Then crawl the website (homepage + about + services + top 3 ad landing pages, parallel WebFetch) and merge into the schema. See references/business-context.md for the full crawl procedure.

Always ask the user (it's faster than guessing): differentiators, competitors, seasonality. Ask for everything else only if the data + crawl can't answer it.

Phase 5 — Personas

Discover 2–3 personas from search terms, top keywords, ad-group themes, landing pages, geo, and device split — all from the dataset already in memory. Persist to {data_dir}/personas/{accountId}.json. Each persona must be grounded in 5+ actual search terms; if not, drop it. See references/persona-discovery.md.

Phase 6 — Report

Lead with the verdict, then the top 3 actions (with dollar impact when possible), then the scorecard, then evidence for dimensions scoring 0–2 only. Cite specific campaigns, keywords, and dollar amounts. Cap at ~80 lines.

End with a single closing line after the handoff to /ads:

*Your audit history is saved to your AdsAgent account — view it at https://adsagent.org.*

Guardrails

1. Read-only skill. Diagnose; don't mutate. Every fix routes through /ads (or /ads-copy, /ads-landing). End the report with one handoff tied to the #1 action.
2. STOP condition — if conversion tracking scores 0–1, recommend pausing spend until it's fixed before recommending anything else. Everything downstream is meaningless without measurement.
3. Always persist business-context.json and personas/{accountId}.json even if the report itself is short — downstream skills depend on them.
4. Name names. Every finding cites specific campaigns, keywords, search terms, and dollar amounts. "Some keywords are underperforming" is not a finding.