TE
TagEasy

AI co-pilot

TagEasy includes three AI-powered features that speed up setup and debugging. They run on either OpenAI or Anthropic, depending on which API key is configured.

Three things the AI co-pilot does

  1. Suggest events — looks at your site, recommends what to track
  2. Normalize taxonomy — renames inconsistent events to the standard convention
  3. Debug events — explains why a specific event isn't firing and how to fix it

1. Suggest events

From your website's Events tab, click + AI suggest. TagEasy:

  1. Fetches your homepage HTML (server-side, SSRF-safe)
  2. Sends it to the AI with the existing list of tracked events
  3. Returns 5-10 high-confidence suggestions: which CTAs to track, which forms, which page-load events, with selectors and proposed event names
  4. You pick which ones to apply → TagEasy creates real VisualEvents

This works best on conversion-driven pages — landing pages, product pages, signup flows. For very large or single-page-app sites, you may want to point it at a specific URL using the optional url parameter.

2. Normalize taxonomy

Useful when you've been adding events ad-hoc and event names have drifted away from the standard. From the Events tab → click Normalize taxonomy:

  1. TagEasy sends the current event list to the AI
  2. AI proposes renames that match {platform}:{provider}:{event}:{websiteSlug}
  3. You see a diff: old name → new name, per event
  4. Accept → TagEasy renames + regenerates GTM tags

Heads up: renaming resets verification history for each renamed event. The new name needs to start firing fresh.

3. Debug events

From the Health tab → click any event →Debug with AI. TagEasy:

  1. Gathers the event metadata + the last 30 days of health logs + verification stats
  2. Asks the AI: likely cause? severity? specific fix steps?
  3. Returns a structured response: likelyCause, severity (low/med/high), fixSteps (1-3 bullets), gtmChanges (what to update in GTM)

Common debug outputs: "selector no longer matches", "page URL changed", "trigger filter too narrow", "GTM tag paused".

Rate limits

Each org can make AI_RATE_LIMIT_PER_ORG_PER_HOUR AI calls per hour (default 30). When you hit the limit, calls return 429 with a retry-after window.

Privacy + data handling

  • The co-pilot sends event metadata + HTML snippets to the configured AI provider — not user PII, login state, or customer data
  • HTML fetching is SSRF-safe — only public URLs from your domain are allowed
  • Calls are logged in AiAuditLog with the input hash + token counts, never the full prompt content
  • If both OPENAI_API_KEY and ANTHROPIC_API_KEY are unset, AI features are silently disabled (the buttons gracefully no-op)

Switching providers

Set the AI_PROVIDER env var to either:

  • openai — uses OPENAI_API_KEY, model defaults to gpt-4o-mini
  • anthropic — uses ANTHROPIC_API_KEY, model defaults to claude-3-haiku

Both providers use the same prompt template, so switching between them produces near-identical output. Pick whichever you already have an account with.

What the AI is NOT

  • It can't directly edit your GTM container
  • It can't see your real user analytics data
  • It can't guarantee that suggested events are the right ones for your business — review before applying

Treat it as a knowledgeable assistant, not a tracking expert who knows your goals. The Health system + verification script + AI co-pilot together cover most setup + maintenance needs.