Track AI referral traffic in GA4 with GTM

#01What GTM’s job is here

GTM ensures the Google tag fires on landing pages with a usable document referrer. Problems GTM can cause:

• A tag that sets fields incorrectly and blanks referrer context.
• Aggressive cross-domain or redirect chains that drop referrer.
• Self-referral issues from subdomain misconfig.
• Accidentally listing chatgpt.com as an unwanted referral while cleaning payment gateways.

Problems GA4 must solve (not GTM):

• Default channel group mapping (AI Assistant channel regex lesson).
• Custom channel group design (custom channel group for AI referrals).
• Explorations and monthly routines (AI search monthly measurement).

#02Practical GTM QA for AI landings

1. From a desktop browser, open an AI assistant, ask a query that surfaces your brand (or paste your URL in a context that sets referrer — note some in-app browsers alter referrers).
2. Land on your site with GTM Preview on.
3. Inspect the first page_view / Google tag hit: is referrer present in variables / collect payload?
4. In GA4 DebugView / Realtime, check Session source/medium.
5. If referrer is empty, test a plain external link from another site to isolate AI-webview quirks vs GTM bugs.

Caveats. Some AI apps open links in ways that send incomplete referrers. That is an ecosystem limit, not always fixable in GTM. Still ensure you are not making it worse.

#03Optional: classify in the data layer

If product wants an event parameter for downstream BigQuery, engineering can set:

dataLayer.push({
  event: 'page_view_enriched',
  ai_referrer: true
});

only when document.referrer matches a maintained hostname list. Keep the list in one Constant Regex / Lookup. Do not invent UTMs that overwrite real campaign data.

Prefer GA4 channel rules as the primary reporting layer so marketers without BigQuery see AI traffic too.

#04Unwanted referrals review

Open Google tag settings → List unwanted referrals (and GA4’s unwanted referrals). Ensure AI hostnames are not listed unless you have an exotic reason. Payment processors and auth hosts often should be listed; AI assistants usually should not.

#05Mentions vs referral traffic

Referral hits mean someone clicked through. Brand mentions inside an AI answer without a click will not appear as GA4 sessions. That is why MentionX monitors answers separately from analytics. Use ChatGPT referral tracking thinking alongside this lesson.

#06Worked example: Ledgerline

Ledgerline is a fictional accounts-payable software company. Details are illustrative.

Ledgerline’s agency had added a broad unwanted referral regex that unintentionally included chatgpt.com during a cleanup. AI-assisted landings looked like Direct. Removing the hostname and building the GA4 custom channel group showed a small but rising share of demo requests from AI click-throughs — enough to justify a content experiment on comparison pages.

Walkthrough

1. Review unwanted referral lists in Google tag + GA4.
2. QA a landing with a known referrer.
3. Implement or verify GA4 AI Assistant / custom channel group using the linked lessons.
4. Add AI referral sessions to the Monday metrics note (n8n reporting can include the segment later).
5. Document hostnames you monitor; revisit quarterly as new AI properties launch.

#07Building the hostname watchlist

Start with: chatgpt.com, chat.openai.com (legacy patterns if still seen), perplexity.ai, gemini.google.com, copilot.microsoft.com, claude.ai — then review your own Referral report for others. The list will change; schedule a quarterly refresh with the AI measurement routine.

#08Landing page experience

If AI referrers bounce at 90%, the issue may be content mismatch, not GTM. Use Explorations to see landing pages for AI channels, then fix pages. Measurement’s job is visibility; CRO’s job is conversion.

#09Mentions workflow with MentionX

Use MentionX (or your monitor) for answer-level visibility; use GA4 for click-through sessions; use CRM for qualified pipeline. Trying to force all three into one GTM event creates nonsense. Keep the layers separate and report them side by side.

#10What GTM can and cannot do here

Can: ensure page_view and conversion events fire so sessions exist; pass useful campaign params when you control the link; avoid breaking referrers with careless redirects.

Cannot: force ChatGPT to send a pretty UTM; invent brand mentions as traffic; fix zero-click answers where the user never visits.

Most of the AI referral work lives in GA4 Admin (channel groups, explorations) and in content/AEO — GTM is the plumbing that must not be broken.

#11Channel group design sketch

Create rules that match referral source / source hostname containing:

• chatgpt.com → channel AI ChatGPT
• perplexity.ai → channel AI Perplexity
• gemini.google.com → channel AI Gemini
• claude.ai → channel AI Claude
• catch-all AI Other for new hosts you add quarterly

Order rules above generic Referral. Test with GA4 custom channel groups guidance in the AI referrals lesson.

#12Reporting cadence

Weekly: sessions and key events from AI channels (expect volatility). Monthly: landing pages and assisted conversions; content gaps for MentionX prompts that drive visits. Quarterly: refresh hostname list; prune dead hosts; brief content on pages that attract AI visits but do not convert.

#13UTMs when you do control the link

If you publish a public page that ChatGPT or Perplexity may cite, you still rarely control their outbound link format. Where you do control links (custom GPTs with actions, plugin surfaces, email from AI-assisted sequences, in-product assistants), use disciplined UTMs (utm_source=chatgpt or a documented standard) and keep them consistent with channel rules so you do not double-count as both Referral and Paid.

#14Mentions without clicks

AEO success often shows as mentions and citations before sessions move. Report MentionX visibility and GA4 AI sessions as related-but-different KPIs. Demanding GTM invent sessions for zero-click answers trains the organisation to distrust dashboards.

#15Exploration recipes that help

1. Free form: rows = Session source / hostname (or your custom channel), columns = Week, values = Sessions + key events. Filter to AI channels.
2. Funnel: Landing page → generate_lead for AI-channel sessions only — finds pages that get cited but do not convert.
3. Path exploration: AI channel landings → next page — shows whether users hit pricing or bounce to blog archives.

Export a monthly CSV for content planning. Pair with MentionX prompt coverage: if a prompt ranks you but the landing page converts poorly, fix the page — do not blame GTM.

#16GTM hygiene that protects referrers

• Avoid redirect rules that force all traffic through a tracking domain that drops referrers.
• Be careful with meta referrer policies that strip to origin or no-referrer sitewide — you may lose AI hostnames.
• Do not overwrite document.referrer in custom scripts.

If SEO/AEO and engineering change referrer policy for "privacy", loop measurement in before launch or you will silently blind AI channel reporting.

#17Try it yourself

Pull last 28 days of Referral hostnames from GA4. Highlight any AI properties. Confirm channel grouping. Fix GTM/Google tag misconfig if referrer was dropped.

#18Quick quiz

1. Should chatgpt.com be an unwanted referral?

Answer: Usually no — you want to see that traffic.

2. Can GTM alone create the AI Assistant channel?

Answer: No. Configure channel groups in GA4.

3. Mentions without clicks appear as sessions?

Answer: No. Sessions need a visit; mentions need separate monitoring.

#19What should you check before finishing the track?

• Unwanted referral list reviewed.
• GA4 channel approach in place or ticketed.
• AI referral note added to monthly measurement routine.

Previous: Audit a GTM container. Or return to the GTM for Marketers track. Related: GA4 AI Assistant channel, Custom channel group, ChatGPT referral traffic.

#20Common questions

Does GTM create the channel? No — GA4 does.

Which hostnames? Maintain a list including chatgpt.com and perplexity.ai.

Fake UTMs in GTM? Don’t overwrite natural referrers.