AI referral traffic in GA4: how to measure visits from ChatGPT, Perplexity, Gemini and Copilot

If an assistant recommends you and the customer clicks through, where does that visit show up? In GA4, by default, it is lumped into "Referral" alongside every other link on the web, or into "Direct" if the assistant stripped the referrer. Either way you cannot see it, which means you cannot report it, which means nobody will fund the work that produced it.

This post gives you a custom channel group that isolates AI referrals, a regex you can paste, and an honest note on what GA4 cannot see.

Which referrers to look for

AI assistants that send traffic with a referrer currently include, as hostnames you will see in GA4's session source:

chatgpt.com
chat.openai.com          (older links; still appears)
perplexity.ai
www.perplexity.ai
gemini.google.com
copilot.microsoft.com
claude.ai
www.bing.com             (only when the landing path or parameters indicate Copilot; see below)

Google AI Overviews and AI Mode are the awkward ones. Clicks from an Overview arrive with google as the source and organic as the medium, indistinguishable from a classic organic click. There is no referrer-level signal. Search Console reports them within the same Web Search totals. Treat them as organic and measure AI Overview presence separately, with a tracker, rather than from analytics.

Step 1: Create a custom channel group

GA4 lets you define your own channel groups without touching the default one. Go to Admin, Data display, Channel groups, Create new channel group. Add a channel at the top of the order (rules are evaluated top down) named "AI Assistants" with this condition:

Session source  matches regex  ^(www\.)?(chatgpt\.com|chat\.openai\.com|perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com|claude\.ai)$

If you want to be generous and catch subdomains and variants:

Session source  matches regex  (chatgpt|openai|perplexity|gemini\.google|copilot\.microsoft|claude\.ai)

The second version will also match openai.com developer documentation links and similar, which is usually acceptable for a small site and worth tightening for a large one.

Leave the rest of the channels as they are. Save. The new group is available in standard reports via the channel dimension picker, and you can set it as the primary group if you like.

Step 2: Build a report

In Explore, create a free-form exploration:

• Dimension: Session source, Landing page + query string
• Metrics: Sessions, Engaged sessions, Key events (conversions), Total revenue if e-commerce
• Filter: Session default channel group (or your custom group) exactly matches "AI Assistants"

Segment by landing page. The pages assistants send people to are the pages they are citing, which is a direct read on which content is working. Compare engagement rate with organic search; the industry pattern is that AI referrals are smaller in volume but often more engaged, because the visitor arrives with a specific expectation set by the answer.

Step 3: Tag your own links where you can

You cannot add UTM parameters to links an assistant generates. You can add them to links you control that assistants might copy: the URLs in your llms.txt, in directory profiles, in structured data url fields. This is of limited value and can cause canonical confusion if done carelessly, so most sites should skip it. The channel group is the reliable method.

What GA4 cannot see

Be straightforward about the gaps when you report.

• Answers without clicks. Most AI answers do not result in a click. The customer reads the recommendation and phones you, or searches your name on Google (which shows as organic or direct). Referral traffic understates AI influence badly.
• Stripped referrers. Some assistants open links in a way that drops the referrer, so the visit lands in Direct. Apps on mobile are the usual culprit.
• AI Overviews and AI Mode. Indistinguishable from organic, as above.
• Copilot via Bing. Some Copilot clicks arrive as bing / organic or bing.com / referral. There is no clean separator.
• Consent. Under UK GDPR, visitors who decline analytics cookies are not in GA4 at all. If your consent rate is 60%, every number here is 60% of reality, and the proportion may differ by source.

The practical conclusion is that GA4 referral data is a lower bound on AI influence, useful for trend and for landing-page insight, and not a measure of how often you are recommended. For that you need to look at the answers themselves.

Pairing analytics with visibility tracking

The two measurements answer different questions:

Question	Source
How often am I recommended?	Visibility tracking (sampled answers per engine)
Which of my pages do assistants cite?	Visibility tracking (citations) and GA4 landing pages
Do AI visitors convert?	GA4 (key events by channel)
Is AI referral traffic growing?	GA4 (sessions by channel over time)
Are assistants getting my facts right?	Visibility tracking (accuracy alerts)

Seoptist's Growth plan connects to GA4 and Search Console with Google OAuth and reports AI referral sessions as a separate segment next to your visibility and share of voice, so the "we were mentioned more" line and the "we got more visits" line sit on the same chart. Plans start at £49 a month; GA4 and Search Console integration is included from Growth at £149. See pricing.

Quick checklist

[ ] Custom channel group "AI Assistants" created with the regex above, placed first in order
[ ] Exploration saved: AI sessions by landing page, with engagement and key events
[ ] Monthly comparison of AI Assistants vs Organic Search: sessions, engagement rate, conversion rate
[ ] Note in every report: GA4 sees clicks only; AI Overviews count as organic; consent rate applies
[ ] Visibility tracking in place for the recommendations that never become clicks