How to See AI Referral Traffic in GA4 (and Why It Undercounts)

Build the segment

AI assistants that send users to your site usually arrive as a referral with a recognizable source domain. The ones worth capturing today include chatgpt.com (and the older chat.openai.com), perplexity.ai, claude.ai, gemini.google.com, copilot.microsoft.com, and the Bing and Google paths that carry AI-surfaced clicks when they are distinguishable.

In GA4, open Explore, create a free-form report, and add a segment with the condition "session source matches regex" and a pattern like:

chatgpt|openai|perplexity|claude\.ai|gemini\.google|copilot\.microsoft

Apply the segment, add sessions, engaged sessions, and key events as metrics, and break down by session source and landing page. Save it. That is the AI referral report. If you want it in the standard reports instead, create a custom channel group with the same regex and name the channel "AI Assistants."

Check the regex against your own referral list first. Sources vary by region and by how the assistant opens links, and new ones appear. Review the list quarterly.

What the number tells you

The landing page breakdown is the useful part. It shows which pages the engines are actually sending people to, which is a direct read on which pages are being cited. Pages with AI referrals and no organic traffic are a sign that an engine likes them for a question Google does not rank them for. Pages with strong citation counts in your tracking and zero referrals are a sign that the citation is being shown but not clicked, which is normal for many question types.

Engagement tends to be high on this traffic. People who click out of an AI answer usually know what they want. In our testing, engaged session rate and conversion rate on AI referrals tend to run above the site average, which is a point worth making when someone asks why this channel deserves attention at a small volume.

Why it undercounts

The GA4 number is a floor, not a measurement of AI influence. Several things keep it low.

Most answers do not produce a click. A user who reads "Acme is a good fit for small accounting firms" and then types "Acme" into Google shows up as organic brand search, not as an AI referral. The engine did the work and got no credit.

Some assistants strip the referrer. Depending on the app, the device, and how the link opens, the session can land as direct traffic with no source at all. Desktop apps and in-app browsers are the usual culprits.

Google's AI surfaces are not separable. A click from an AI Overview or from AI Mode arrives as google / organic. There is no flag in GA4 or Search Console that tells you it came from the AI layer.

Consent and ad blockers remove a share of all sessions, and this channel's users tend to be technical and tend to block more.

What to pair it with

Because the referral number is a floor, pair it with two other signals. First, branded search volume, which tends to rise when mentions rise even when clicks do not. Second, direct measurement of the answers themselves: run a fixed prompt set across the engines and record mentions and citations. The referral report tells you what happened after a click. The prompt tracking tells you what the engines are saying whether or not anyone clicks. At AEO GEO Labs the weekly report carries both side by side, and the pattern is consistent: share of answer moves first, branded search follows, and referrals trail last and smallest.

Ten minutes, then a habit

Build the segment today. Look at the landing pages, not just the total. Note which pages are drawing clicks and compare that list with the pages you think are your best. The mismatch is usually instructive. Then stop treating the referral number as the measure of AI visibility and start treating it as one corroborating signal. If you want the answer-side measurement set up alongside it, AEO GEO Labs builds that as the first step of every engagement.