Google Analytics Now Has an "AI Assistant" Channel. Here's What It Still Doesn't Count

On May 13, 2026, Google Analytics 4 (GA4) started sorting some visits into a new built-in channel called "AI Assistant." For many teams, that line in the acquisition report will quickly become "our AI traffic." It is a useful number. But it counts only one slice of how AI assistants affect a brand.

The channel works by checking where a visit came from. That design leaves out three things: clicks from Google's own AI answers, assistant clicks that arrive without a source attached, and the many AI answers that mention a brand without producing any click at all. This article explains each gap, using Google's and OpenAI's documentation and the few published measurements, and then shows how to read the number.

What Google actually added

GA4 now labels a visit as AI traffic when its referrer matches a recognized AI assistant. The referrer is the piece of information a browser passes along to say which site the visitor just came from. When it matches, GA4 sets the medium to "ai-assistant," the campaign to "(ai-assistant)," and the channel to "AI Assistant." No setup is needed.

Google's launch note names ChatGPT, Gemini and Claude. The default channel group definitions name ChatGPT, Gemini, Deepseek, Copilot and Grok as example sources. Google does not publish the full list on that page, and Perplexity is not named there. If Perplexity matters to you, check whether its visits show up under AI Assistant or under Referral in your own reports.

Because the rule depends on the referrer, the useful question is what a referrer-based rule cannot see. There are three answers.

Blind spot 1: Google's own AI answers are not in it

The channel definition says it "excludes Google's AI Overviews and AI Mode." AI Overviews are the AI summaries at the top of Google results; AI Mode is Google's conversational search. Clicks from both stay in the Organic Search channel, mixed in with clicks from classic blue links.

Google Search Console does not separate them either. Google's guide to AI features says traffic from AI Overviews and AI Mode is reported in the Performance report "within the 'Web' search type." The Performance report help page confirms that a click on a link in an AI Overview counts as an ordinary click. So neither tool gives you a separate line for Google's AI surfaces.

The Conductor 2026 AEO/GEO Benchmarks Report ran into the same wall. Because Google Analytics groups AI Mode, AI Overviews and organic Google traffic together, Conductor left AI Mode out of its AI referral figures.

What this means: if Google's AI answers drive a meaningful share of your AI-influenced visits, the AI Assistant channel will not show it.

That covers Google. For non-Google assistants, the channel has a different weakness: it needs the referrer to arrive intact.

Blind spot 2: visits without a referrer land in Direct

Some visits from AI assistants reach your site with no referrer. In practitioner testing by Seer Interactive, when a visit carries neither a referrer nor UTM tags, GA4 records it as Direct. UTM tags are labels added to the end of a link, such as utm_source=chatgpt.com, that tell analytics where a click came from. Seer notes that in-app browsers and copy-pasted links "will still land in direct."

ChatGPT adds utm_source=chatgpt.com to many of the links it sends, as OpenAI's publisher FAQ and third-party testing report. Seer's testing found this happens most reliably on cited sources, not every time. For example, when a user asks ChatGPT for a specific URL, it typically returns the link without UTM tags.

How often does the referrer go missing? Nobody knows reliably. Two vendors have published figures:

• Clickport found that 35.7% of the AI search sessions it could recognize had no referrer. Those sessions were identified only because they still carried a source tag, so visits with neither signal are not in the count. The figure came from a 30-day snapshot of a self-selected group of sites.
• Loamly, an analytics vendor, reported that about 70.6% of the AI visits it tracked arrived without a referrer, according to a review of the figure by David Rosam.

The two numbers come from small samples measured in different ways, and they are far apart. The safe conclusion is that referrer loss is real, but its size is unknown and will differ by site.

Be careful with how these figures get repeated. Both describe the share of AI visits that lost their referrer. Neither measures the share of your Direct traffic that came from AI. As Rosam puts it: "So when someone tells you 70% of your direct traffic is AI, that is not what anybody measured."

Even if every AI click were tracked perfectly, though, a bigger gap would remain.

Blind spot 3: the answer that never produces a click

Much of what AI assistants do for a brand never becomes a visit. Conductor's benchmark of 1,215 enterprise domains in the US found that AI referrals made up about 1.08% of all website traffic. That is roughly 1 visit in 100. The figure leaves out Google's AI Mode, and if it relies on referrers like GA4 does, it is likely an undercount. Still, it is small.

Meanwhile, assistants regularly read pages and name brands in their answers. OpenAI's crawler documentation describes ChatGPT-User, an agent that fetches pages "for certain user actions in ChatGPT and Custom GPTs." That fetch is a request from OpenAI's servers, not a person's browser running your analytics tag. So the assistant can read your page, use it in an answer, and recommend you, without GA4 recording a session.

What this means: a brand can be recommended often in AI answers and still see a small AI Assistant number. GA4 measures the click. It cannot measure the recommendation that came before it, or the one that never led to a click. We can't put a precise number on that unseen share, but given how small AI referral traffic is, it is likely to be most of the picture.

So the channel is real data with clear limits. The next question is how to use it.

How to read the AI Assistant number

Treat it as a floor. Seer describes the channel as "best read as a floor on your AI traffic, not a ceiling." It counts clicks from recognized assistants that kept their referrer. Actual AI-driven click traffic is at least this high, and probably higher.

Use it as a trend line from May 2026 on. Google's launch note says nothing about older data, and Seer reports that the channel is forward-only: GA4 won't reclassify past sessions. Comparing this channel with months before the launch will mislead you. If you need history, a custom channel group you already maintain is the better source.

Keep your own custom channel group. The built-in list is not published in full, and new assistants appear often. Seer recommends keeping a custom grouping and reviewing your GA4 data monthly for new AI referrers. In practice, the Referral report is a good place to spot assistants the default list misses.

Audit Direct traffic, but don't relabel it. Seer recommends auditing Direct for hidden AI visits, and Rosam suggests two checks. First, look for deep, hard-to-guess pages that receive Direct visits with no earlier page on your site before them. Rosam calls this "a signature of an AI referral. A signature, not a measurement." Second, add a "how did you hear about us" field to forms. These checks give you clues, not a count.

Measure visibility in the answers themselves. The biggest blind spot, influence without a click, can only be seen by looking at what assistants say. That means running the prompts your buyers ask and tracking how often your brand is mentioned, recommended and cited. Generative engine optimization (GEO) platforms are built to track brand presence and citations across assistants. Whatever you use, report answer-level results alongside GA4 rather than instead of it. Our guide to mentions versus recommendations covers how to present those metrics to leadership.

The bottom line

GA4's AI Assistant channel answers one narrow question well: how many visitors clicked through from a recognized non-Google assistant with the referrer intact. It does not answer how visible your brand is in AI answers. Google's AI Overviews and AI Mode sit inside Organic Search, lost referrers sit inside Direct, and much of AI's influence, probably most of it, never produces a click. Read the channel as a lower bound on AI click traffic. Then measure the answers directly, because that is where influence without a click shows up.