How to Track AI Traffic in GA4 and What It Leaves Out

How to track AI referral traffic in GA4 with one custom AI channel that combines Google's AI Assistant channel and a tested regex, plus what GA4's count leaves out and why.

Angelique Swain · · 7 min read

Cover reading How to Track AI Traffic in GA4 and What It Misses, in cream type on a dark Firemark ground
On this page

Someone in the leadership meeting will ask how much traffic AI sends you, and GA4 is where you'll go for the number.

Knowing how to track AI traffic in GA4 gets you most of the way, because Google now files part of it under its own channel. The rest takes one custom channel group and a regex you can check line by line.

Even then, AI referral traffic in GA4 is a partial measure of AI-driven visits, and the reasons are documented.

Does GA4 track AI traffic already?

GA4 files visits from AI assistants under a default AI Assistant channel when the referrer matches Google's list, a change Google announced on May 13, 2026.

Google's release note names ChatGPT, Gemini and Claude. Its channel definitions page names ChatGPT, Gemini, Deepseek, Copilot and Grok as example sources. A matching visit gets the medium "ai-assistant" and the campaign "(ai-assistant)".

To see it, open Reports > Acquisition > Traffic acquisition. The first column shows your primary channel group, which follows the default group unless someone changed it in Admin, and AI Assistant appears as its own row once matching visits arrive.

Three limits come with it. Google doesn't publish the full list of assistants it recognizes, so you can't check what's on it.

The channel also excludes Google's AI Overviews and AI Mode, which sit in Organic Search. And the release note doesn't say whether sessions from before May 13 were reclassified, so older AI visits may still sit under Referral.

How to track AI traffic in GA4

One custom AI channel that takes both Google's AI Assistant traffic and your own regex gives you a single row, and Google applies custom groups to past data.

You'll need the Editor role or above on the property. Then follow Google's channel group steps:

  1. Go to Admin, and under Data display click Channel groups.
  2. Click Create new channel group. GA4 starts you with a copy of the default group.
  3. Name the group, for example "Channels with AI".
  4. Click Add new channel and name it "AI assistants".
  5. Click Add condition group and add two conditions joined by OR: Default channel group exactly matches "AI Assistant", or Source matches regex, with the regex from the next section.
  6. Click Save channel, then Reorder and drag "AI assistants" above Google's built-in AI Assistant channel, Referral and any other channel it should override. Click Apply.
  7. Click Save group.

The OR condition matters. Without it you get two rows, your "AI assistants" and Google's "AI Assistant", and a report filtered to one of them leaves out the visits only the other caught. If your editor won't join the two conditions with OR, keep Google's channel directly below yours and filter on both rows.

GA4 puts each visit in the first channel it matches, top to bottom, so the order decides the count. A standard property allows two custom channel groups of up to 50 channels each, and custom groups aren't included in the BigQuery export.

To read the result, go back to Traffic acquisition and switch the first column to your new group. "AI assistants" now holds the sessions Google files as AI Assistant plus those whose source matches your regex, including sessions from before May 13.

Which regex catches AI referrers?

A short list of assistant hostnames, anchored at both ends, catches the main AI referrers without sweeping in look-alike sites.

^(.*\.)?(chatgpt\.com|chat\.openai\.com|perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com|copilot\.com|copilot\.cloud\.microsoft|m365\.cloud\.microsoft|claude\.ai|deepseek\.com|grok\.com|meta\.ai|chat\.mistral\.ai)$

Three details make it behave. GA4 regex is a full match by default, so the pattern has to describe the whole source value.

Channel definitions ignore case, so "ChatGPT.com" and "chatgpt.com" land in the same row. And (.*\.)? allows a prefix such as www. or chat., but only before a dot, so a host like notchatgpt.com can't slip in.

Which hosts does the regex match?

The regex matches the listed assistant domains and their subdomains, and rejects Google Search, OpenAI's developer site and look-alike domains, as the table shows.

Which hosts does the regex match?
Source in GA4AssistantMatches?
chatgpt.comChatGPTYes
chat.openai.comChatGPT (old host, now redirects)Yes
www.perplexity.ai or perplexity.aiPerplexityYes
gemini.google.comGeminiYes
copilot.microsoft.com or copilot.comMicrosoft CopilotYes
m365.cloud.microsoft or copilot.cloud.microsoftMicrosoft 365 Copilot (work accounts)Yes
claude.aiClaudeYes
chat.deepseek.comDeepSeekYes
grok.comGrokYes
www.meta.aiMeta AIYes
chat.mistral.aiLe Chat (Mistral)Yes
google.comGoogle Search, including AI Overviews and AI ModeNo, by design
platform.openai.comOpenAI's developer siteNo

Two hosts have moved: on November 20, 2026, chat.openai.com redirected to chatgpt.com and copilot.microsoft.com redirected to copilot.com. A list written before those moves can miss Copilot visits.

A match proves less than it seems. The pattern says what a source looks like, not that an assistant actually sends that hostname as its referrer, and we didn't test each assistant's outbound referrals.

Three gaps remain by design. The prefix also admits developer hosts such as docs.perplexity.ai. A visit from a Microsoft 365 hostname doesn't prove the referral came from a Copilot answer. And Grok used inside X arrives as x.com, which GA4 files under Organic Social.

Google's help page gives an example regex too. Its first branch, ^.*ai, matches any source that ends in "ai", which includes every site on a .ai domain that links to you. Plenty of those are startups that have never answered a question in their lives. A named list takes longer to type and far less time to audit.

Once a month, sort the Referral rows by sessions and look for an assistant host you haven't listed. Google's own note says to update the regex when the assistants you measure change.

How do you see AI landing pages?

A free-form exploration filtered to your AI channel shows which pages assistants send people to, and how those visits engage.

  1. Click Explore in the left menu and create a Blank exploration.
  2. Under Variables, add the session-scoped dimension for the custom group you created, such as Session Channels with AI, plus Session source and Landing page + query string.
  3. Add the metrics Sessions, Engaged sessions and Key events.
  4. Drag the landing page and source into Rows and the metrics into Values.
  5. Add a filter: that same dimension (Session Channels with AI in this example) exactly matches "AI assistants".

Check your data retention first. Explorations read event data, which GA4 keeps for two months by default, and a standard property can raise that to 14 months under Admin > Data collection and modification > Data retention. Raising it doesn't bring back data already deleted, and standard reports aren't affected by the setting.

What does GA4's AI count leave out?

GA4's AI channel captures identifiable visits from AI assistants. Some AI-driven visits lose their source information, while Google's AI search clicks sit under Organic Search. Answers that never produce a visit need a separate visibility measure.

Answers that never produce a visit

An answer someone reads without clicking isn't a visit GA4 missed. It's visibility outside GA4's scope, and it's common: people rarely click through from Google's AI summaries. Pew Research Center studied the browsing of 900 US adults in March 2025. On Google searches that showed an AI summary, people clicked a regular result in 8% of visits, against 15% when no summary appeared. A link inside the summary got a click in 1% of those visits.

Bar chart from Pew: people clicked a regular result on 15% of Google visits with no AI summary and 8% when a summary appeared; a link inside the summary got 1%
Pew Research Center, July 2025: share of Google visits with a click, 900 US adults, March 2025

Those are findings from one study of Google searches in March 2025, not current click rates for every engine. Still, a visit that never happens leaves nothing for GA4 to record, however often the answer named you.

Google's channel definitions put clicks from AI Overviews and AI Mode in Organic Search, alongside ordinary search results. That's a classification inside GA4, not traffic missing from it. Those visits arrive from google.com, so a source regex can't separate them from other Google clicks.

Search Console's Generative AI performance reports show impressions for your URLs in those two features, which covers part of the gap. The report may not appear for a site without enough impressions. Firemark's guide to tracking Google's AI answers sets the two views side by side.

Visits that lose the referrer

If a visit arrives without a referrer or campaign information, GA4 may report it as Direct or attribute it to an earlier identifiable source, since its session-scoped source dimensions use a last-click model for paid and organic channels. Google lists typed URLs, links opened from PDFs or Word files, redirects that drop tags and ad blockers among the causes of Direct traffic.

AI answers feed several of those paths. A reader who copies a URL out of a chat window and pastes it into the browser arrives as typed traffic. If the page holding the link sets the Referrer-Policy header to no-referrer, the browser sends no referrer at all. Some ChatGPT visits arrive tagged utm_source=chatgpt.com, which GA4 reads as the source even when the referrer is missing.

Don't read a rise in Direct as hidden AI traffic, though. GA4 alone can't establish whether an untagged, referrer-less visit originated in an AI answer, so the AI share of Direct is unknown.

What does an answer count add?

An answer count samples what AI engines say about your brand when asked a fixed set of buyer questions, whether or not anyone clicks.

Firemark asks those questions on the engines in your plan, drawn from ChatGPT, Perplexity, Google AI Overviews, Google AI Mode and Gemini. For each answer, it records whether your brand is named, whether your site is linked and which sites are cited. Firemark measures answers generated for your tracked questions. It doesn't measure how many real buyers saw those answers, and it doesn't track the clicks, visits or sales that follow, so GA4 stays the place for those.

What does an answer count add?
SourceWhat it countsWhat it can't see
GA4 AI channelVisits arriving with an AI referrer or tagAnswers read without a click; AI Overviews and AI Mode clicks
Search Console Generative AI reportsImpressions of your URLs in AI Overviews and AI ModeChatGPT, Perplexity and Gemini; brand names without a link
Answer count (Firemark)Tracked answers naming your brand, out of all answers collected; links reported separatelyHow many real buyers saw the answers; clicks, visits and sales

Report the three side by side, each labeled with what it counts. Firemark's guide to GEO KPIs for client reports shows a one-page layout with GA4 visits labeled as a partial measure.

Frequently asked questions

What does utm_source=chatgpt.com mean?

utm_source=chatgpt.com is a campaign tag on a link clicked inside ChatGPT, and GA4 reads it as the session source. A Source rule like the regex above then files the visit as AI traffic, even when the referrer is missing.

Does GA4 count clicks from AI Overviews?

GA4 counts clicks from AI Overviews and AI Mode as Organic Search, by Google's channel definitions. They can't be split from ordinary Google results inside GA4, so use Search Console's Generative AI reports for impressions.

Why does ChatGPT traffic show as direct?

ChatGPT traffic can show as direct when the visit arrives with no referrer and no tag, for example a URL copied from a chat and pasted into the browser. GA4 may also credit such a visit to an earlier source the user came from, and it can't tell either way that the visit started in ChatGPT.

Can GA4 show which prompt sent a visitor?

GA4 can't show the prompt or the answer behind a visit, only the source and the landing page. To see what the answers said, track brand mentions in ChatGPT and other engines, which is the half GA4 doesn't hold.

How often should you update the regex?

Update the regex whenever an assistant changes its host or a new one starts sending visits, and check monthly. Google's help page gives the same advice, and Copilot's move to copilot.com shows why.

Count the visits and the answers

GA4 can now tell you how much AI traffic arrives, and a tested regex fills the gaps in its list. The number is a partial measure of AI-driven visits, and many AI answers never produce a visit at all. Firemark counts how often answers to your tracked questions name and link you, and its method page lists what it leaves out.

Angelique Swain AI-first SEO Strategist

Written by

Angelique Swain · AI-First SEO Strategist

Angelique Swain is an AI-first SEO strategist with over a decade in organic search across retail, medical and B2B software. Most of her work now is on how AI engines choose which brands to name. She builds fixed prompt sets, re-runs them to see how much the answers vary, and traces what the cited pages have in common. At Firemark she writes about measuring AI visibility and what to do once you can see it.

  • Cover reading Unlinked Brand Mentions in AI Answers: What They Mean, in cream type on a dark Firemark ground

    Measure AI visibility

    What Do Unlinked Brand Mentions Mean in AI Answers?

    An unlinked brand mention in an AI answer is an answer that names your brand and cites none of your pages. It happens when the answer cites nothing or cites other sites, and only the second points at pages you can work on.

    · 6 min read

Find out if AI recommends you

Start tracking your brand. New accounts get one free run.