Google Ads audit

Score an account in six areas, in the order they usually waste money, and end with a fix list. The audit only reads.

When to use

A full account review: new client takeover, quarterly health check, "why is this account underperforming". For one campaign before launch use the pre-launch-check skill; for "what changed last week" use performance-insights.

Inputs

• account: the profile alias (default profile if not given).
• Period: last 30 days unless the user names one. GAQL has no LAST_90_DAYS; for longer periods use segments.date BETWEEN 'YYYY-MM-DD' AND 'YYYY-MM-DD'.

Steps

Call account_profile first (currency, linked GA4). Call guide("gotchas") once before writing GAQL. Money in raw run_gaql_query output is already converted from micros.

1. Measurement (25 points). guide("measurement").
   • What each campaign bids on: run_gaql_query with SELECT campaign.name, conversion_goal_campaign_config.goal_config_level, conversion_goal_campaign_config.custom_conversion_goal FROM conversion_goal_campaign_config and SELECT campaign.name, campaign_conversion_goal.category, campaign_conversion_goal.origin, campaign_conversion_goal.biddable FROM campaign_conversion_goal WHERE campaign_conversion_goal.biddable = TRUE. A campaign with campaign-level goals ignores the account defaults.
   • Which actions record: SELECT segments.conversion_action_name, metrics.all_conversions, metrics.conversions FROM customer WHERE segments.date DURING LAST_30_DAYS.
   • If GA4 is linked: ga4_conversions for the same period. Big gaps between GA4 lead events and Ads conversions mean tracking or goal setup is broken, not "no leads".
2. Location leaks (15).
   • SELECT campaign.name, campaign.geo_target_type_setting.positive_geo_target_type FROM campaign WHERE campaign.status = 'ENABLED'. Anything other than PRESENCE is a finding.
   • Spend outside the targets: SELECT campaign.name, user_location_view.country_criterion_id, user_location_view.targeting_location, metrics.cost_micros, metrics.clicks FROM user_location_view WHERE segments.date DURING LAST_30_DAYS. Rows with targeting_location = false are spend from people outside the target areas (country level).
   • Interest-based spend at any level: SELECT campaign.name, geographic_view.country_criterion_id, geographic_view.location_type, metrics.cost_micros, metrics.clicks FROM geographic_view WHERE segments.date DURING LAST_30_DAYS. location_type = AREA_OF_INTEREST is spend from people who were only interested in the targets.
3. Search terms and negatives (20).
   • Top terms by cost (get_search_terms sorts by clicks, so use GAQL here): SELECT search_term_view.search_term, campaign.name, metrics.cost_micros, metrics.clicks, metrics.conversions FROM search_term_view WHERE segments.date DURING LAST_30_DAYS ORDER BY metrics.cost_micros DESC LIMIT 50. Flag irrelevant or competitor terms with spend and no conversions.
   • Negative coverage: SELECT shared_set.name, shared_set.member_count FROM shared_set WHERE shared_set.type = 'NEGATIVE_KEYWORDS' and SELECT campaign.name, campaign_criterion.keyword.text FROM campaign_criterion WHERE campaign_criterion.negative = TRUE AND campaign_criterion.type = 'KEYWORD'. A shared list only counts where it is attached: SELECT campaign.name, shared_set.name FROM campaign_shared_set WHERE shared_set.type = 'NEGATIVE_KEYWORDS' AND campaign_shared_set.status = 'ENABLED'.
4. Structure (10).
   • get_adgroup_performance. Dynamic Search Ads groups (ad_group.type = 'SEARCH_DYNAMIC_ADS') and broad-only groups that match the same searches as another campaign leak traffic between audiences.
   • get_keyword_performance: the same keyword in several campaigns competes with itself.
5. Budgets and bidding (15).
   • get_campaign_budgets and SELECT campaign.name, campaign.primary_status, campaign.primary_status_reasons, campaign.bidding_strategy_type, metrics.search_budget_lost_impression_share, metrics.search_rank_lost_impression_share FROM campaign WHERE campaign.status = 'ENABLED' AND segments.date DURING LAST_30_DAYS. Budget-limited campaigns with good cost per conversion are opportunities; learning or limited statuses need time, not changes.
6. Ad quality and policy (15). guide("policy").
   • Quality Score and its components: SELECT campaign.name, ad_group.name, ad_group_criterion.keyword.text, ad_group_criterion.quality_info.quality_score, ad_group_criterion.quality_info.creative_quality_score, ad_group_criterion.quality_info.post_click_quality_score, ad_group_criterion.quality_info.search_predicted_ctr, metrics.cost_micros FROM keyword_view WHERE ad_group_criterion.status = 'ENABLED' AND segments.date DURING LAST_30_DAYS. Weight low scores by cost: a low score on a keyword with no spend doesn't matter.
   • SELECT campaign.name, ad_group.name, ad_group_ad.ad.id, ad_group_ad.ad_strength, ad_group_ad.policy_summary.approval_status FROM ad_group_ad WHERE ad_group_ad.status = 'ENABLED'.
   • Asset coverage (sitelinks, callouts, images) at all three levels, because account-level assets serve every campaign: SELECT campaign.name, campaign_asset.field_type FROM campaign_asset WHERE campaign_asset.status = 'ENABLED', SELECT customer_asset.field_type FROM customer_asset WHERE customer_asset.status = 'ENABLED' and SELECT campaign.name, ad_group.name, ad_group_asset.field_type FROM ad_group_asset WHERE ad_group_asset.status = 'ENABLED'.

Score each area from its findings: full points with none, minus points proportional to the money or reach at stake.

Output

# Google Ads audit: <alias> (<period>)
Score: NN/100

| Area | Score | Top finding |
|---|---|---|
| Measurement | x/25 | ... |
...

## Findings
1. <finding> — <evidence with numbers> — money at stake: <currency amount/month>
...

## Fix list (highest impact first)
1. <fix> — tool: `ads_add_negative_keywords` / `ads_update_networks` / ... — effort: low/medium/high

Money uses the profile currency. Say which areas could not be checked (for example no GA4 linked) instead of guessing.

Never

• Never call a write tool during the audit. Fixes are listed, not applied.
• If the user then asks to apply a fix, run the write tool without confirm first (preview), show it, and apply only after an explicit yes. If the profile is read-only (the write tools are missing), tell the user to run npx google-marketing-agent accounts set <alias> writes true and restart their MCP client. Never edit config files or work around it.
• Never present benchmark ranges as targets for this account; use guide("benchmarks") to explain how to read them.