An AI agent for Google Tag Manager

What it reads

With read access the agent sees the published container the way an analyst would: every tag, trigger, variable and folder, with all their settings. It starts with Google's read-only Tag Manager permission and asks for nothing more until you raise the client's level.

• Tags, triggers, variables and their settings
• Consent settings on every tag
• The published version, and every new version that goes live
• Search across the whole container by name or value

What the GTM audit checks

The GTM audit is a fixed set of checks with no model involved, so the same container always gives the same findings. Each finding says what is wrong, why it matters and who fixes it: the agent, you, or your developer.

• Tags that fire twice or never
• Tags that ignore the visitor's consent choice
• Duplicated Google Ads conversions and missing Conversion Linkers
• GA4 event names that break GA4's rules, and debug mode left on
• Naming, paused and orphaned items, and Universal Analytics leftovers

Checked in a real browser

A container can look right and still send the wrong thing. The site check loads your key pages as a visitor would, once with the cookie banner accepted and once refused, and reads what the site really sends. It runs in a browser on our side, or in your own Chrome through the extension.

• The dataLayer pushes on each page
• Every hit to GA4, Google Ads and Meta
• Consent Mode defaults and updates
• Tags that fire before the visitor has chosen

What it builds

On the Read & write level the agent creates and edits tags, triggers and variables in its own workspace, so nothing mixes with your team's work and nothing goes live. To track a new event, you can pick the element or do the action on the site in your own browser, and the agent builds the trigger, the variables and the GA4 event tag from that.

• New events from a sentence, such as tracking a form or a button
• Fixes from the audit, one workspace per change
• A version ready for you to look at in Preview

What it publishes

On the Publish level the agent can put a version live, and only after a person approves it in the chat. Every change is logged, and the previous version is one step away. When the new version adds a GA4 event, the task watches GA4's realtime report for it while you have the task open.

Google Tag Manager automation with limits you set

Owners and admins set each client to Read, Read & write or Publish. The level is enforced three times: in the permissions asked of Google, in the tools the agent is given, and in a check each time a tool runs. A request in the chat cannot raise it.

• Read: the agent audits and reports
• Read & write: it builds in its own workspace
• Publish: it puts a version live after approval
• Autopilot, if an owner or admin turns it on: low-risk builds go live on their own

Watched after it goes live

Tag Manager tells nobody when a version is published, so monitoring looks. The published container is checked on a schedule set by your monitoring plan, from once a day to every few hours, and more often for a while after a new version. A new version is compared with the one before and audited, and an alert is raised unless it was published from Tracking Whisper.