Gtm Audit

GTM Audit

A one-time audit of a Google Tag Manager container. GTM rarely fails loudly — it fails as a duplicate GA4 tag double-counting conversions, a tag with no trigger that never fires, or analytics tags firing before consent. This skill reads the container and finds those.

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

Works with

Prerequisites Check

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

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

1. Sign up at https://cogny.com and connect Google Tag Manager (GA4 and your ad platforms
   optional, to verify IDs and event names).
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: the container is only readable through the GTM API.

If Cogny is connected but mcp__cogny__mcp__Google_Tag_Manager__tool_list_containers isn't, say "connect Google Tag Manager in Cogny" and stop.

Health check: call list_integrations, then tool_list_accounts as the cheap read (and one cheap read per platform used in Step 6b, e.g. tool_list_ad_accounts). health_status misses revoked tokens, so a 401, Forbidden or scope error is 🔴 "reconnect "; don't retry it. If a tool named here is missing, call refresh_mcp_discovery once; new tools only load after the client reconnects, so mark that check ⚠ "needs reconnect" for this run.

Usage

/gtm-audit — full container audit /gtm-audit tags — tag health only /gtm-audit triggers — trigger coverage only /gtm-audit consent — Consent Mode wiring only

Steps

1. Account & container discovery

Copy code
tool_list_accounts
tool_list_containers   (account_id)

If multiple containers exist, ask which to audit. Note the container type (web vs server) — this audit targets web containers. Then tool_list_workspaces (account_id, container_id) for the workspace_id every later call needs (use the Default Workspace unless the user names another).

2. Tag inventory

Copy code
tool_list_tags         (account_id, container_id, workspace_id)

For tags that need detail, call tool_get_tag. Bucket every tag by type (GA4 config, GA4 event, Google Ads conversion, Google Ads remarketing, Floodlight, custom HTML, third-party).

Flag:

• Paused tags still sitting in the container (dead weight, or accidentally off)
• More than one GA4 configuration tag for the same Measurement ID — the classic double-counting bug
• Multiple tags sending the same event to the same destination
• Custom HTML tags doing what a native tag template should do (fragile, slow)
• Google Ads conversion tags with no conversion label / ID set

3. Trigger coverage

Copy code
tool_list_triggers     (account_id, container_id, workspace_id)

Cross-reference every tag's firingTriggerId against the trigger list.

Flag:

• Tags with no firing trigger — they will never fire (orphaned)
• Triggers attached to no tag — dead triggers
• All Pages triggers on conversion or event tags (should fire on a specific event)
• Click / form triggers with no conditions (fire on every click — noisy, wrong data)

4. Variable hygiene

Copy code
tool_list_variables    (account_id, container_id, workspace_id)

Flag:

• Data Layer Variables referenced by tags but not defined
• Defined variables referenced by nothing (clutter)
• No version of the consent-state variables if Consent Mode is expected

5. Consent Mode v2

Copy code
tool_audit_container_privacy   (account_id, container_id, workspace_id)

Start from its list of tags without consent triggers, third-party trackers and cookie variables, then inspect tags and the container setup for Consent Mode wiring.

The privacy tool only counts GA4 tag types, so it misses pixels pasted into Custom HTML. Also:

• Scan every Custom HTML tag (type: html; the code is in the html parameter, so tool_get_tag if tool_list_tags didn't return it) for fbq( (Meta), twq( (X), ttq. (TikTok), lintrk( / _linkedin_partner_id (LinkedIn). Count each as a third-party tracker.
• Read every tag's consentSettings.consentStatus: needed (with the required consent types) is wired, notNeeded is an explicit opt-out, and notSet / missing means no additional consent check. A marketing pixel with notSet and an All Pages trigger is a 🔴 consent finding.

Flag:

• Analytics / Ads tags firing with no consent check — a compliance and data risk in the EEA and UK
• No consent initialization (default denied state) before tags load
• ad_storage / analytics_storage / ad_user_data / ad_personalization not all handled

6. GA4 wiring correctness

For each GA4 event tag (tool_get_tag):

Flag:

• Event tags not referencing the GA4 config tag (or no config tag at all)
• Event names that break GA4 rules — uppercase, spaces, reserved prefixes, >40 chars (cross-reference the /ga4-events conventions)
• Ecommerce event tags not reading the ecommerce object from the data layer
• Missing parameters on key conversion events (value, currency, transaction_id)

6b. IDs against the connected platforms (enhanced when connected)

Call list_integrations once. For each healthy platform, read the IDs it actually owns and compare with the IDs hard-coded in GTM tags:

Flag:

• 🔴 A tag sending to an ID the platform doesn't list (old property, deleted pixel, another client's account): data goes nowhere or somewhere wrong. Exception, GA4: a Measurement ID missing from tool_list_data_streams is 🟡 "not listed or not accessible", not 🔴. The connected Google login may simply lack access to that property; ask the user to confirm it
• 🟡 A platform conversion action (status ENABLED) with no GTM tag behind it, unless it is imported from GA4 or sent server-side. Google Ads actions whose conversion_action.type starts with GOOGLE_ANALYTICS_4_ are GA4 imports: they never have an AW- tag, so don't flag them "no AW- tag" (check them in Step 6c instead)
• 🟡 Segment list_destinations shows a GA4 or ad-platform destination and GTM fires the same tag: likely double counting. Ask which path is meant to be live.

Platforms that aren't connected, or whose read returned any API error (401, scope, quota, GAQL error): say "IDs not verified" for their tags. An error is never evidence that an ID is wrong.

6c. Event names match key events and Ads imports (case-sensitive)

Collect the event name each GA4 event tag sends (eventName parameter). When eventName is {{Event}} (the built-in variable that forwards the dataLayer event), the tag's own config holds no name: take the names from its firing triggers instead (firingTriggerId → tool_get_trigger, type customEvent, the arg1 value of each condition in customEventFilter, e.g. FirstLogin; with a regex match, list each alternative). Skipping this misses the first_login vs FirstLogin mismatch below. Compare each name, exactly and case-sensitively, with:

• GA4 tool_list_conversion_events (property_id): every key event's eventName
• Google Ads GA4 imports from Step 6b: actions with type GOOGLE_ANALYTICS_4_CUSTOM / GOOGLE_ANALYTICS_4_PURCHASE (the import name ends in the GA4 event name)

Flag:

• 🔴 A key event or Ads import whose name matches a GTM event only when case or separators are ignored (e.g. GTM sends first_login, GA4 key event is FirstLogin). GA4 event names are case-sensitive, so the key event never fires and the Ads import stays at zero. This was the root cause of dead key events in a live run.
• 🟡 A key event with no GTM tag sending it at all (unless it's an automatic or enhanced-measurement event, or sent from code / server-side; ask)

7. Naming & workspace status

Copy code
tool_get_workspace_status   (account_id, container_id, workspace_id)

Flag:

• Unpublished changes sitting in the workspace — live site is behind the container
• No naming convention (GA4 - Event - purchase style) — unmaintainable at scale
• Many open workspaces (merge-conflict risk)

8. Score and report

Copy code
GTM Audit — [Container Name]
Container Health Score: X/100

Score breakdown:
  Tag health ......... X/25   (duplicates, orphans, paused)
  Trigger coverage ... X/20
  GA4 wiring ......... X/20   (incl. key-event name matches)
  Consent Mode ....... X/20
  Hygiene ............ X/15   (naming, variables, workspace)

Inventory: X tags · X triggers · X variables  ·  Unpublished changes: [yes/no]

🔴 Critical   — duplicate GA4 config, tags firing before consent
🟡 Important  — orphaned tags, bad event names, missing parameters
🟢 Optimization — naming, unused variables, custom HTML to native

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

9. Record findings

jsonCopy code
{
  "title": "Two GA4 configuration tags for G-XXXX — every page view counted twice",
  "body": "'GA4 - Config' and 'GA4 Base (old)' both load Measurement ID G-XXXX on All Pages. GA4 is counting every page_view and session twice, inflating traffic and deflating conversion rate. Pause the older tag and republish.",
  "action_type": "tag_fix",
  "expected_outcome": "Accurate GA4 traffic and conversion-rate numbers",
  "estimated_impact_usd": 0,
  "priority": "critical"
}

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

Action types: tag_fix, trigger_fix, consent_configuration, container_hygiene, ga4_wiring.

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. Duplicate GA4 config tags and pre-consent firing are always critical — they corrupt data and create compliance exposure.
2. An orphaned tag fires nothing — say which tags are dead and why.
3. Quote real numbers: tag/trigger/variable counts, the specific tag names.
4. Read-only. Never edit or publish the container — surface and recommend. Fixes go through /gtm-setup, where every write asks first.
5. Only call an ID wrong when the platform was read. Unverified is not wrong.