How to Measure AI Referral Traffic in GA4 (2026)

AI referral traffic in Google Analytics 4 means the visits GA4 records with an AI assistant, such as chatgpt.com, perplexity.ai, claude.ai or gemini.google.com, as their source. It is worth measuring, but only one step of four. A session shows that someone arrived from an AI site. It does not show which answer they read, whether that answer named you, or whether the visit turned into a buyer.

We built this method while checking our own AI referrals at Everything Design. The first filter we wrote quietly missed some of them, for a reason that will catch anyone who uses GA4’s regular expressions. This post sets out the 4 measures, the filter that works, and the checks that tell a real enquiry from a visit.

What does an AI referral session actually tell you?

An AI referral session tells you that a visit reached your site with an AI assistant recorded as its source. It is the third of 4 separate measures, and a count at one step proves nothing about the next.

Measure	What it records	What it cannot prove
Mention	An AI answer names your company	That it linked to you or sent anyone
Citation	An AI answer links to one of your pages as a source	That anyone clicked the link
Referral session	GA4 logs a visit with an AI site as its source	Which answer the visitor saw, or why they clicked
Key event	The visitor did something you marked as important, such as sending the contact form	That the enquiry was a buyer you would want

Mentions and citations come from an AI visibility tool that runs prompts and reads the answers. Sessions and key events come from GA4. The two sets of data are not linked, so a rise in both is not evidence that one caused the other until you can trace an enquiry back through the chain.

Why does a GA4 filter miss AI referrals?

A GA4 filter usually misses AI referrals because GA4’s “matches regex” needs the expression to match the whole value, not part of it. Google’s own help page says GA4 regex is full match and case sensitive by default. A filter of chatgpt|perplexity|gemini therefore matches a source that is exactly “perplexity”, and skips one recorded as “chatgpt.com”.

That is what happened to us. Our first report returned a handful of sessions from the sources GA4 had recorded as bare names, and none from the ones recorded as domains. Nothing in the report warned us. The numbers simply looked small.

The fix is to wrap the list so it can match anywhere in the value:

.*(chatgpt|openai|perplexity|claude|anthropic|gemini|copilot|deepseek|grok|mistral).*

Before you trust the result, compare it with an unfiltered list of every session source over the same dates and read down it by eye. Any AI source you can see in the full list and not in the filtered one is a gap in the expression.

Which AI visits never show up as referrals?

Visits that arrive without a referrer cannot be attributed to an AI site at all, and GA4 files them under Direct. That can happen when someone copies a link out of an answer, or opens it from an app that does not pass on where the click came from. No filter recovers those visits.

So treat the AI referral count as a floor, not a total. Report it with that caveat attached, and do not compare it month on month without checking that the way assistants pass on clicks has not changed in between.

Where should you look once the filter works?

Look at the landing pages, not the total. Add the landing page as a second dimension to the filtered report and ask 2 questions of each page that receives AI visits.

First, does the page answer the question that probably brought the visitor there, in its opening lines? A page an AI assistant cites is a page a reader arrives at halfway through a decision, so it has to make sense on its own.

Second, does the page lead somewhere useful next? An article that gets AI visits should link to the work that proves its claim and to a way to get in touch. We set out why AI answers often describe a B2B company wrongly, and how to fix it, in why AI search gets your B2B company wrong.

How do you tell a qualified enquiry from a visit?

Decide what counts before you look. A key event such as a form submission is a starting point, not a result. An enquiry counts as qualified only when the submission names a need you serve and your own records confirm the company fits the customers you work with.

Then compare AI-referred enquiries with enquiries from search and from referrals, on that same definition. If AI visitors send fewer enquiries but better ones, that is worth knowing. If they send none, the pages they land on are the problem to fix, not the AI visibility.

What should a B2B marketing team do this quarter?

Run the 4 steps in order. Fix the regex and check it against the unfiltered source list. Add landing pages to the report. Rewrite the opening of every page that gets AI visits so it answers its question in the first 2 sentences, and link each one to proof and to a contact route. Then set a qualified-enquiry definition and review it monthly.

None of this needs a new tool. It needs a report you can trust, and the discipline not to call a session a lead.

FAQs

Does GA4 have a channel for AI traffic?
Check your own property’s default channel groups, because Google changes them. If there is none, create a custom channel group with the regex above, so AI visits stop being split between Referral and Unassigned.

Are AI referral sessions leads?
No. A session is a visit. It becomes a lead only when the visitor does something you defined in advance, and a qualified lead only when that enquiry fits your customer.

Why do AI mentions and AI visits not move together?
Because an answer can name a company without linking to it, and a link can appear without anyone clicking it. Track the 2 separately and do not read one as evidence for the other.

How often should we check?
Monthly is enough for most B2B sites, alongside a quarterly read of which pages get AI visits and whether they lead anywhere.

If your pages get AI visits and no enquiries, book a call. We will look at where the visits land.