Pmax Audit

Performance Max Audit

A deep audit of your Performance Max campaigns. PMax hides most of its levers — this skill pulls the ones the API still exposes (asset groups, asset coverage, search themes, audience signals, exclusions, channel mix) and flags what's leaking spend.

Requires: Cogny Agent subscription ($9/mo) — Sign up

Works with

Step 0: Cogny check

If no mcp__cogny__* tools exist (no list_integrations, no cogny_status), print this and stop:

Copy code
Performance Max Audit needs your marketing data through Cogny's MCP gateway.

1. Sign up at https://cogny.com and connect Google Ads (plus Microsoft Ads, GA4, Shopify or
   WooCommerce if you use them).
2. Add Cogny to .mcp.json in your project:
   {"mcpServers": {"cogny": {"type": "http", "url": "https://app.cogny.com/mcp"}}}
3. Restart Claude Code; it opens a browser for the OAuth login on first use.

Nothing to run without Cogny.

If Cogny is connected but mcp__cogny__mcp__Google_Ads__tool_execute_gaql is missing, say "connect Google Ads in Cogny" (or "reconnect Google Ads in Cogny" if it's listed but unhealthy or a read returns 401/permission denied) and stop. If only Microsoft Ads is connected, run step 1's Microsoft branch and say that the asset-level audit needs Google Ads.

Usage

/pmax-audit — audit every Performance Max campaign in the account /pmax-audit Holiday Retail — audit one campaign by name or id

Why PMax needs its own audit

PMax is a black box compared to Search. There are no keywords and no per-placement bids. The few real levers — asset group ad strength, asset coverage, search themes, audience signals, and exclusions — are what this skill inspects, because they are the only things you can actually change.

Steps

Google Ads money fields are micros (divide by 1,000,000). An empty GAQL result (no rows, or the tool returns no output) means zero: no matching entities or no activity in the window. Report it as 0, not as an error. Query errors come back as explicit error messages.

1. Find the Performance Max campaigns

sqlCopy code
SELECT campaign.id, campaign.name, campaign.status,
  campaign.bidding_strategy_type, campaign.maximize_conversion_value.target_roas,
  campaign.target_cpa.target_cpa_micros, campaign_budget.amount_micros,
  metrics.cost_micros, metrics.conversions, metrics.conversions_value
FROM campaign
WHERE campaign.advertising_channel_type = 'PERFORMANCE_MAX'
  AND segments.date DURING LAST_30_DAYS
ORDER BY metrics.cost_micros DESC

Microsoft Ads (if connected): tool_list_accounts for the account_id, then tool_get_campaigns (account_id, campaign_types: ["PerformanceMax"]), then tool_get_report (account_id, report_type: "campaign", campaign_ids, start_date, end_date). Report spend, conversions and revenue as a separate block; Microsoft PMax is often an import of the Google one, so compare CPA/ROAS between the two and flag a gap over 2x.

2. Asset group health

sqlCopy code
SELECT asset_group.id, asset_group.name, asset_group.status,
  asset_group.ad_strength, campaign.name,
  metrics.cost_micros, metrics.conversions, metrics.conversions_value
FROM asset_group
WHERE campaign.advertising_channel_type = 'PERFORMANCE_MAX'
  AND segments.date DURING LAST_30_DAYS

Flag:

• Ad strength POOR or AVERAGE on any spending asset group
• Campaigns with a single asset group (no audience/theme segmentation)
• Asset groups with status REMOVED/PAUSED still configured but campaign starved

3. Asset coverage per group

asset_group_asset.performance_label was removed in API v25 (UNRECOGNIZED_FIELD). Use primary_status / primary_status_reasons plus the asset's own metrics:

sqlCopy code
SELECT asset_group.name, asset_group_asset.field_type, asset_group_asset.status,
  asset_group_asset.primary_status, asset_group_asset.primary_status_reasons,
  asset.id, asset.type,
  metrics.impressions, metrics.clicks, metrics.conversions, metrics.cost_micros
FROM asset_group_asset
WHERE campaign.advertising_channel_type = 'PERFORMANCE_MAX'
  AND segments.date DURING LAST_30_DAYS

Assets with no impressions in the window drop out of this result; run it once more without the metrics and the segments.date filter to count every linked asset.

For each asset group, count assets by field_type and compare to Google's max slots:

Flag:

• Any group below the recommended count for headlines, images, or video
• No video asset — Google auto-generates a low-quality one if you skip it
• Assets whose primary_status is NOT_ELIGIBLE or LIMITED (read primary_status_reasons for why, e.g. a policy limit), and low performers: assets with impressions but a CTR or conversion rate far below the other assets of the same field_type in the group
• Only one image aspect ratio (limits inventory: blocks whole channels)

4. Search themes & audience signals

sqlCopy code
SELECT asset_group.name, asset_group_signal.search_theme.text,
  asset_group_signal.audience.audience
FROM asset_group_signal
WHERE campaign.advertising_channel_type = 'PERFORMANCE_MAX'

Flag:

• Asset groups with no audience signal — PMax starts cold, wastes budget learning
• No search themes set (up to 25 allowed — they steer the Search surface)
• Search themes that overlap your brand terms (inflates brand cannibalization)

5. Search category insight (where spend actually goes)

campaign_search_term_insight accepts one campaign per query. Loop over the campaign ids from step 1:

sqlCopy code
SELECT campaign_search_term_insight.category_label,
  metrics.clicks, metrics.conversions, metrics.conversions_value
FROM campaign_search_term_insight
WHERE segments.date DURING LAST_30_DAYS
  AND campaign_search_term_insight.campaign_id = '<id>'
ORDER BY metrics.clicks DESC
LIMIT 50

An empty category_label is the bucket of uncategorised terms, not an error.

Flag categories with high clicks and zero conversions — candidates for negative keywords (PMax negatives are set at the account level, or via support).

6. Brand exclusions, URL expansion & new customers

campaign.brand_guidelines_enabled is about brand colours and fonts in generated assets, not brand exclusions. Brand exclusions are BRAND_LIST campaign criteria:

sqlCopy code
SELECT campaign.name, campaign_criterion.type, campaign_criterion.negative,
  campaign_criterion.brand_list.shared_set
FROM campaign_criterion
WHERE campaign_criterion.type = 'BRAND_LIST'
  AND campaign.advertising_channel_type = 'PERFORMANCE_MAX'

A PMax campaign with no row here has no brand exclusion. Final URL expansion and other automation settings:

sqlCopy code
SELECT campaign.name, campaign.asset_automation_settings
FROM campaign
WHERE campaign.advertising_channel_type = 'PERFORMANCE_MAX'

Look for the final-URL-expansion entry (e.g. FINAL_URL_EXPANSION_TEXT_ASSET_AUTOMATION) with asset_automation_status: OPTED_IN. campaign.customer_acquisition_setting.optimization_mode is not recognised by the API, so read the new-customer split from conversions instead. This segment accepts conversion metrics only (adding metrics.cost_micros fails):

sqlCopy code
SELECT campaign.name, segments.new_versus_returning_customers,
  metrics.conversions, metrics.conversions_value
FROM campaign
WHERE campaign.advertising_channel_type = 'PERFORMANCE_MAX'
  AND segments.date DURING LAST_30_DAYS

An empty result means 0 conversions in either segment. If step 1 shows conversions for the campaign but this returns nothing, new-vs-returning isn't being measured (no customer lists or new-customer goal), so say "not measured" rather than "0 new customers".

Flag:

• No brand exclusion list — PMax happily spends on your own brand searches
• Most conversions value from returning customers, or no new-vs-returning data at all (new-customer goal likely not configured), for accounts that want growth, not just remarketing-flavored conversions
• Final URL expansion ON with no URL exclusions (PMax can send traffic anywhere on the domain)

7. Listing groups (retail PMax only)

If the campaign has a Merchant Center feed:

sqlCopy code
SELECT asset_group.name, asset_group_listing_group_filter.type,
  asset_group_listing_group_filter.case_value.product_type.value
FROM asset_group_listing_group_filter
WHERE campaign.advertising_channel_type = 'PERFORMANCE_MAX'

Flag everything dumped into one "All products" node — no way to prioritize high-margin SKUs. Recommend /shopping-campaign-audit for full feed analysis.

7b. Enhanced when connected: is the ROAS real?

PMax claims credit for brand searches and returning buyers. If a store is connected, pull paid revenue for the same 30 days (Shopify tool_list_orders with query_filter: "created_at:>=YYYY-MM-DD financial_status:paid"; WooCommerce get_sales_report with date_min, date_max) and show PMax conversions_value as a share of it. With GA4, tool_list_account_summaries for the property, then tool_run_report (dimensions: ["sessionGoogleAdsCampaignName", "newVsReturning"], metrics: ["sessions", "keyEvents", "purchaseRevenue"]; keep sessions, as GA4 drops rows where every metric is 0) shows how much PMax revenue comes from returning users. A high returning share plus no brand exclusion makes step 6 critical.

8. Score and report

Copy code
Performance Max Audit — [Account Name]
PMax Score: X/100  ·  30-day spend: [X]  ·  Conv: [X]  ·  ROAS: [X.Xx]

Per campaign / asset group:
| Campaign | Asset group | Ad strength | Signal? | Video? | Spend | ROAS |
|----------|-------------|-------------|---------|--------|-------|------|

🔴 Critical   — broken signals, POOR ad strength, no brand exclusion
🟡 Important  — thin assets, missing video, no search themes
🟢 Optimization — add aspect ratios, expand headlines

Top 3 Actions:
1. [Highest $ impact]
2. ...
3. ...

9. Record findings

jsonCopy code
{
  "title": "PMax 'Holiday Retail' has no audience signal — slow, expensive learning",
  "body": "Asset group 'Gifts' spends $90/day with no audience signal attached. PMax is exploring blind. Add a signal from converting customers (customer match list or your top-converting segment) to speed up learning.",
  "action_type": "audience_refinement",
  "expected_outcome": "Faster learning, lower CPA within 2-3 weeks",
  "estimated_impact_usd": 1200,
  "priority": "high"
}

Cloud uses create_ticket with the same content, fields mapped per section 2 of the routing guide.

Action types: asset_optimization, audience_refinement, negative_keyword, campaign_optimization, bidding_strategy, brand_exclusion.

Recording findings

Use create_ticket when the workspace has it (Cloud) and create_finding otherwise (Solo). The fields differ; see section 2 of ../marketing-engineer/references/mcp-routing.md.

Critical rules

1. Asset strength and audience signals are the real levers. Spend the audit there, not on metrics you can't influence.
2. Always check brand exclusions. Unexcluded brand traffic is the most common PMax money leak.
3. Quote real numbers — ad strength label, asset counts, category clicks.
4. Read-only. Surface issues; never edit asset groups. Write tools such as tool_add_asset_to_asset_group are used only on the user's request, and the skill asks first.
5. Platform and store numbers stay side by side, each in its own currency.