Event audit

A read-only hygiene pass over the three systems that count events: what fires, what GA4 receives, and what Google Ads bids on. It produces a findings table with evidence and a fix list for the gtm-build skill. For "Ads says 0 but we got leads", start with tracking-audit instead.

When to use

• Suspected double counting, or GA4 event lists full of near-duplicates.
• After a site or agency change, or quarterly as routine hygiene.
• Before trusting conversion data for a bidding change.

Inputs

• account with gtmContainerId and ga4PropertyId set (check with account_profile; the setup skill adds them).
• The tracking plan or target matrix if one exists (from tracking-plan): it defines what "should" fire.

Steps

Call account_profile first. Use guide("gtm") for the rules a container should follow and guide("measurement") for counting and freshness. Everything here is read-only.

1. Tag Manager. gtm_overview, then gtm_list with entity: "tags", "triggers" and "variables", gtm_builtin_variables with action: "list", and gtm_get on each conversion, GA4 and custom HTML tag. Check:
   • tags with no firing trigger (they never run);
   • paused tags still referenced: another tag's setup/teardown sequencing points at them, or the plan still counts on them for a key event or primary Ads action;
   • Ads conversion tags but no Conversion Linker (no gclidw tag on All Pages and no Google tag configured with the AW- ID): Safari/ITP conversions go unattributed;
   • Ads conversion tags sending a constant 0 value (it overrides the action's default value);
   • duplicates: two tags with the same event name, conversion label or measurement ID on the same trigger;
   • conversions on page views: an Ads conversion tag on All Pages, or on a Page View / DOM Ready trigger limited to a confirmation path (a fragile success signal);
   • text or element-visibility triggers (fragile: they break silently when copy or language changes);
   • GA4/Ads pairs for the same event with different trigger lists;
   • variables referenced as {{Name}} that exist neither in the variables list nor in the enabled built-ins;
   • consent notSet where the site serves consent regions: fine on Google tags in advanced Consent Mode when a consent default exists (they read it themselves); a finding when there is no consent default, or on vendor (non-Google) tags (details: consent-audit).
2. GA4. ga4_list_key_events, ga4_custom_dimensions with action: "list", then ga4_custom_report with dimensions: ["eventName"], metrics: ["eventCount"], for the 30 days ending 3 days ago (dateFrom 32 days ago, dateTo 3 days ago; the last 2 days are still processing). Check:
   • key events with no events in the period (unused, or their tag is broken);
   • events the GTM tags send that never appear in the report;
   • near-duplicate names that differ by case, underscores or plural (Generate_Lead, generate-lead, generate_leads): GA4 treats them as separate events;
   • parameters mapped in GA4 event tags (gtm_get → event settings) that aren't registered custom dimensions.
3. Google Ads. run_gaql_query:
   • All actions (zero-conversion actions don't return rows in a date-segmented query): SELECT conversion_action.id, conversion_action.name, conversion_action.category, conversion_action.primary_for_goal, conversion_action.status FROM conversion_action WHERE conversion_action.status = 'ENABLED'.
   • Last 30 days: SELECT conversion_action.id, conversion_action.name, conversion_action.category, conversion_action.primary_for_goal, conversion_action.status, metrics.all_conversions FROM conversion_action WHERE segments.date DURING LAST_30_DAYS.
   • Goals: SELECT custom_conversion_goal.id, custom_conversion_goal.name, custom_conversion_goal.conversion_actions FROM custom_conversion_goal. Check: enabled actions with no conversions in 30 days, primary actions with category PAGE_VIEW, actions in no custom goal and not primary (nothing bids on them), two primary actions that count the same lead.
4. Cross-check. Ads counts only conversions after an ad click, dated to the click; GA4 counts every lead from every source, dated to the event. Compare like with like: ga4_custom_report with dimensions: ["eventName", "sessionSourceMedium"] (or sessionGoogleAdsCampaignName), metrics: ["eventCount"], same window, keeping only the google / cpc rows, against Ads conversions for the matching action. Leads that came after a click late in the window can still be landing in Ads (click date vs event date), so expect some lag at the edges. Only on that filtered comparison does a gap of more than about 20% point at different triggers, consent or a broken label; never compare Ads with all GA4 leads.

5. Findings table. One row per finding, worst first:

   Severity	Where	Finding	Evidence	Fix (tool)
   high	GTM	Ads conversion tag fires on All Pages	tag 12 "Ads - Lead" firingTriggerId 2147479553	pause with gtm_upsert; rebuild on a lead trigger with gtm_build
   high	GTM	Ads conversion tags but no Conversion Linker	3 awct tags; no gclidw tag, no Google tag with an AW- ID	create it with gtm_upsert (via gtm-build)
   medium	GA4	form_name sent but not registered	GA4 tag 15 maps form_name; not in custom dimensions	ga4_custom_dimensions create
   low	Ads	"Old Lead" no conversions in 30 days, secondary	0 all_conversions	leave, or rename with ads_update_conversion_action

   • high: inflates or loses conversions that bidding uses (double counting, conversions on All Pages, broken primary action, ungated vendor pixels in consent regions, notSet Google tags with no consent default).
   • medium: reporting is wrong or fragile (unregistered parameters, text triggers, mismatched pairs, conversions on a page-view trigger limited to a confirmation path).
   • low: clutter (dead tags, unused key events, near-duplicate names with no traffic).
6. Fix list. Group the fixes for the gtm-build skill (GTM changes, custom dimensions, key events) and Google Ads (ads_set_conversion_primary, ads_update_conversion_action, ads_create_custom_goal, ads_assign_custom_goal). Order: high first, and anything bidding relies on before reporting clean-up.

Output

1. The findings table (step 5), with counts per severity.
2. The fix list (step 6), ready to hand to gtm-build.
3. What couldn't be checked and why (for example, no GA4 access, or a container with server-side tagging).

Never

• Never change anything from this skill: fixes go through gtm-build or the Ads write tools after the user agrees.
• Never call an event dead from the last 2 days of GA4 data.
• Never recommend deleting conversion actions with history; make them secondary or rename them.
• Never treat two events as duplicates on name alone without checking their triggers and counts.
• If the profile is read-only, the whole audit is still delivered; tell the user that fixing needs npx google-marketing-agent accounts set <alias> writes true and an MCP client restart.