How to Track AI Search Traffic in GA4 (And What It Misses)

• GA4 now has a native AI Assistant channel — traffic from recognised assistants gets medium ai-assistant.
• Build a custom channel group anyway, so the definition is yours and not Google's to change.
• A large share of assistant clicks arrive with no referrer and land in Direct. That is structural.
• AI Overviews and AI Mode clicks are not separable — they arrive as google / organic. Use Search Console instead.
• Treat every AI referral figure as a floor, and validate it against server logs if the number matters.

The fastest check: is anything arriving at all?

Before configuring anything, look. In GA4, open Reports → Acquisition → Traffic acquisition, set the primary dimension to Session source, and use the search box for the hosts that matter:

Assistant	Source value you will see
ChatGPT	chatgpt.com (older sessions: chat.openai.com)
Perplexity	perplexity.ai
Gemini	gemini.google.com
Copilot	copilot.microsoft.com
Claude	claude.ai

If you find nothing at all, that is information too — but check the date range covers at least a month before drawing conclusions. Assistant referrals on a small site are lumpy, and a fortnight of zero is not evidence of anything.

The native AI Assistant channel

GA4 added an AI Assistant channel to the default channel group. When a session arrives from an assistant GA4 recognises, it is stamped automatically:

• Medium becomes ai-assistant
• Channel group becomes AI Assistant
• Campaign becomes (ai-assistant)

This is a real improvement. Before it existed, assistant traffic was scattered through Referral and had to be dug out by hand every time.

Two caveats keep it from being the whole answer. First, the set of assistants GA4 recognises is Google's to define, and it changes — a tool you care about may or may not be in it this quarter. Second, and more seriously, it can only classify what arrives with a referrer. More on that below, because it is the important part.

Build a custom channel group anyway

The reason to duplicate the work is control. A custom channel group is defined by you, applies to historical data in reports, and does not shift when Google revises its own definitions.

In GA4: Admin → Data display → Channel groups → Create new channel group.

Add a channel — call it AI assistants — with a condition matching Source against a regex covering the hosts you care about:

.*(chatgpt\.com|chat\.openai\.com|perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com|claude\.ai).*

Then do the step people skip: drag the new channel above the built-in Referral rule. GA4 evaluates channel rules in order and stops at the first match. Left at the bottom, your carefully written rule never fires, because Referral already claimed the session — and the report looks exactly like it did before, which is how this mistake survives for months.

Add hosts as new tools appear. That is the maintenance cost of owning the definition, and it is small.

The part most guides leave out

Here is the structural problem, and no GA4 setting solves it.

A large share of assistant clicks arrive with no referrer at all. A link opened from the ChatGPT desktop app, from a mobile app, or from certain in-app browsers reaches your server with an empty Referer header. GA4 receives a session with no origin information, and files it where it files everything with no origin information: Direct.

So your AI referral number is not a measurement of AI-driven visits. It is a measurement of AI-driven visits that happened to carry a referrer. The true figure is higher, and the gap is not a constant — it varies with how your audience uses these tools, which means it also cannot be corrected with a fixed multiplier.

Three things follow:

• Do not present the number as precise. It is a floor with an unknown ceiling.
• Watch the trend, not the level. A floor that triples is still meaningful evidence of growth.
• Watch Direct alongside it. A rise in Direct with no campaign or brand explanation, moving in step with your AI referrals, is consistent with the same underlying traffic.

If the number genuinely matters — say you are deciding whether to allow or block an assistant's crawler — validate it against server logs, where the request arrives whether or not a referrer does. Logs also show you the crawler traffic itself, which never appears in GA4 at all because bots do not execute your analytics tag.

Why AI Overviews and AI Mode are a different problem

A common request is a report separating "traffic from AI Overviews" — and it cannot be built, for a simple reason. A click from an AI Overview is a click from a Google search results page. It arrives as google / organic, exactly like a click from a blue link. Nothing in the referrer distinguishes them.

Search Console is the better instrument here. Impressions and clicks from Google's AI surfaces are included in your Search performance data, so the comparison worth making is not "AI traffic versus normal traffic" but impressions versus clicks over time on the queries you care about. A query whose impressions hold steady while its clicks fall away is the fingerprint of an answer being given above your result. Before concluding that, though, check your average position — impressions without clicks usually means depth, not an Overview.

That measurement problem, and what can honestly be concluded from it, is covered in can you opt out of AI Overviews. The mechanism behind the click loss is in what AI Overviews are and how they work.

What to do with the number once you have it

Reporting is only worth building if it changes a decision. Three it can inform:

Whether to allow an assistant's crawler. If a product sends you real visitors, blocking its indexing crawler has a measurable cost. If it has never sent one, the calculation is different. Which bot does which job is the subject of AI crawlers and robots.txt, and it is about to matter more given Cloudflare's September default block.

Which pages get cited. Landing-page breakdowns within assistant traffic show what these systems actually pull from. It is rarely the pages you would guess, and it is usually the specific, self-contained ones — consistent with what the research on getting cited in AI search finds.

Whether the visitors are worth having. Assistant referrals arrive pre-briefed: the user has already read a summary and clicked anyway. Compare their engagement against organic before deciding what the channel is worth. Sometimes it is better. Occasionally it is much better.

The honest summary: GA4 can now tell you the shape of this, and cannot tell you the size. Build the reporting, watch the direction, and resist quoting the number to two decimal places.

Sources: Google Analytics Help — Custom channel groups · Google Analytics Help — What's new in Google Analytics · Google Search Central — AI features and your website