How to see AI referrals in analytics without fooling yourself
A custom AI channel in GA4 takes ten minutes. Knowing what it misses, from app visits to branded search, is what keeps the number honest.

The short answer
To see AI referrals in analytics, add a GA4 custom channel group with an AI assistants channel above Referral, matching the engines' referrer domains. Treat the result as a floor: app visits, copied links and AI Overview clicks land elsewhere, and many AI-influenced visits arrive later as branded search. Check the channel, branded search and self-reported attribution weekly.
Key takeaways
- In SE Ranking's study of 101,574 websites (January–April 2026), AI search sent 0.32% of visits, so a separate channel is needed or the signal disappears inside Referral.
- In the same study, ChatGPT sent 74.78% of AI referrals and Gemini 11.56%, which is why most AI channel rules start with ChatGPT's domains.
- GA4 assigns a session to the first channel whose rule matches, so the AI channel must sit above Referral in the group.
- Clicks from Google's AI Overviews and AI Mode arrive as organic search and cannot be split out in GA4.
- Read the AI channel together with branded search demand and a "how did you hear about us" field, never on its own.
In this article
Why give AI referrals their own channel?
Because otherwise they vanish into Referral, next to directory listings and partner links. The volumes are small, so they need to be isolated before anyone can see a trend. In SE Ranking's study of 101,574 websites from January to April 2026, AI search accounted for 0.32% of visits.1 SE Ranking sells SEO software, and its sample leans towards sites that use its tools.
The mix of engines is just as uneven. In the same study, ChatGPT sent 74.78% of AI referrals and Gemini 11.56%.2 Other studies put ChatGPT's share anywhere from about half to nearly nine in ten, for reasons we set out in why the referral-share studies disagree.
Our view: a small, separately tracked number beats a large, blended one. The point of the channel is to watch direction and landing pages over months, not to prove a return in week one.
AI referrals, January–April 2026
Small, and dominated by one engine
How to set up the AI channel in GA4
Create a custom channel group, copy the default channels into it, and add one new channel for AI assistants. Google's help page on custom channel groups explains the mechanics: groups apply to reports retroactively, and a session goes to the first channel whose rule it matches.3 That ordering rule is where most setups go wrong.
The ten-minute setup
- Create the group
Admin, Data display, Channel groups. Copy the default group so every existing channel stays.
- Add an AI assistants channel
Condition: session source matches a regular expression listing the engines' referrer domains.
- Move it above Referral
Otherwise Referral claims the sessions first and the new channel stays empty.
- Save and compare
Open the traffic acquisition report with the new group and check that Referral fell by the same amount.
- Review the list monthly
Engines add domains and apps; check the Referral channel for new AI sources.
For the source rule, start with the referrer domains you can see in your own Referral report. A typical list covers chatgpt.com, chat.openai.com, perplexity.ai, gemini.google.com, copilot.microsoft.com and claude.ai. Join them with a vertical bar and escape the dots. Check your Referral report before trusting any list, including this one.
Keep the rule on session source, not on landing-page parameters alone. Some engines tag outbound links and some do not, and tags get stripped when people copy a link. The referrer domain is the more stable signal for the sessions that keep one.
What the channel will never count
A lot. The AI channel counts sessions that arrive with an AI engine as the referrer, and nothing else. Four groups of AI-influenced visits land somewhere else in your reports, and two of them are probably larger than the channel itself.
What the AI channel counts, and where the rest goes
Counted in the AI channel
- Clicks on a cited link in a browser session
- Clicks from engines that pass a referrer
- Visits from tagged links that keep their tag
Lands elsewhere
- AI Overviews and AI Mode clicks: organic search
- Desktop and mobile apps without a referrer: direct
- Copied and pasted links: direct
- Readers who later search your name: organic or paid search
The Google case is structural. Clicks from AI Overviews and AI Mode carry Google as the referrer, so GA4 files them under organic search with every other Google click. If you need to separate them, look at Search Console and at the case for tracking AI Overviews and AI Mode as two channels, not at GA4.
Bing has moved further. In February 2026 Microsoft opened a public preview of an AI Performance report in Bing Webmaster Tools, which reports how a site appears in Copilot and Bing's AI answers.4 It measures citations, not visits, but it covers ground GA4 cannot see.
Why AI-influenced visits arrive as branded search
Because people often read an answer, note a name and search for it later. Similarweb studied whether users shown a brand in ChatGPT went on to visit its site, and through which channel. Search Engine Land reported the study in June 2026.5 Our note on that study covers what it found about the channels those visits came through.
Citation rates help explain why most sectors see so few direct clicks. Similarweb found that the share of ChatGPT answers carrying a citation rose from 1.6% in June 2025 to 6.8% in May 2026, on US data.6 In May 2026 that share was about 23% for travel and under 4% for professional services.6 Similarweb sells traffic intelligence.
Our view: if your sector sits near the bottom of that range, the AI channel will understate your exposure most. A law firm or consultancy can be named in many answers and see almost no referral sessions, while branded search climbs.
A weekly check that keeps the number honest
Read four signals together, every week, in the same order. None is enough alone; the pattern across them is what tells you whether AI exposure is growing.
| Signal | Where | What to look for |
|---|---|---|
| AI channel sessions and landing pages | GA4, custom channel group | Direction over eight weeks; which pages engines send people to |
| Branded search demand | Search Console, queries with your name | Impressions rising while non-brand stays flat |
| Self-reported source | "How did you hear about us" on forms | An "AI assistant" option, with a free-text field |
| Answer presence | A fixed prompt set, run on several days | Whether you are named and which page is cited |
Our weekly check. Illustrative: adjust the window to your traffic volume.
The fourth row is the one analytics cannot replace. At Sigzen AI, our AI answer monitor runs a fixed prompt set across six engines: ChatGPT, Gemini, Perplexity, Claude, Copilot and Google AI. It uses official APIs where offered and a licensed AI-answer data provider otherwise. Audits run each prompt on three separate days, because a single run is noise.
What not to conclude from the AI channel
Do not compare its conversion rate with organic search and call AI traffic better or worse. The sessions it counts are a skewed sample: people who clicked a cited link in a browser. They are not representative of everyone who saw your name in an answer.
Do not report the channel's sessions to a board as "AI traffic" either. Report it as AI referral sessions, a floor, next to branded demand and answer presence. Our guide to which AI search number belongs in a board report covers the wording, and how to tie AI visibility to pipeline covers the next step.
Our view: the honest version of this report is less impressive and far more useful. A team that knows its AI number is a floor makes better decisions than one that thinks it is the whole story.
Sources
- SE Ranking, AI traffic study: 101,574 websites, 250 countries, Jan–Apr 2026 (Jun 2026)
- SE Ranking, AI traffic study: share of AI referrals by engine, Jan–Apr 2026 (Jun 2026)
- Google Analytics Help, custom channel groups
- Microsoft Bing, AI Performance in Bing Webmaster Tools, public preview (Feb 2026)
- Similarweb ChatGPT recommendations study, via Search Engine Land (Jun 2026)
- Similarweb, AI search stats 2026: ChatGPT answers with a citation, US, Jun 2025–May 2026 (Jul 2026)


