AI Traffic in GA4: How to Measure Visits from ChatGPT, Gemini or Perplexity (2026 Guide)

GA4 can finally count AI traffic, provided you know what it does not count

Yes, you can see in Google Analytics 4 the visits coming from ChatGPT, Gemini, Copilot or Perplexity. Since May 2026, GA4 even has a dedicated channel, "AI Assistant", enabled by default. But that channel captures only part of the phenomenon: it excludes AI Overviews, does not name every engine, and a share of AI traffic arrives without a referrer, drowned in direct traffic. This guide gives the complete method: what the native channel covers, the custom configuration that completes it, the numbers that justify the effort, and the right way to read curves that structurally underestimate reality.

Why measure this traffic now

The volume remains modest for most sites, but its trajectory and its quality already make it a channel in its own right. According to Adobe Analytics data covering US retail sites, traffic from generative AI grew 393% year over year in the first quarter of 2026; above all, its quality has flipped: in March 2025 this traffic converted 38% worse than other channels, in March 2026 it converts 42% better, with revenue per visit 37% higher (Adobe data reported by Decrypt). A visitor arriving from an AI answer has already received the context, the comparison and often the recommendation: they are not coming to discover, they are coming to conclude.

Not isolating this traffic therefore means steering a rising channel blind, and depriving your AI visibility work of its quantified proof.

What GA4 now does on its own: the "AI Assistant" channel

Since an update rolled out in mid-May 2026 (reported by PPC Land), GA4 includes an "AI Assistant" channel in its default channel group. How it works, as described in the official Google Analytics documentation: when a session's referrer matches Google's recognized list of AI assistants, GA4 automatically assigns the medium "ai-assistant" and groups the session under this channel. The documentation cites ChatGPT, Gemini, Deepseek, Copilot and Grok as examples.

Two caveats, straight from the same documentation and from what we observe:

• The channel explicitly excludes Google's AI Overviews and AI Mode. A click from an AI Overview is still counted in Organic Search, indistinguishable from a classic click in GA4 (our analysis of the AI Overviews impact details this point and the Search Console side of the reading).
• The full list of recognized assistants is not published. Perplexity, Claude and Le Chat do not appear in the documentation's examples, and the market observes that part of this traffic still lands in Referral. Hence the complementary setup below.

The method in 5 steps

1. Get the lay of the land in two minutes

In Explore, create a free-form exploration with the "Session source" dimension and the "Sessions" metric, then filter the source with the regex chatgpt|openai|perplexity|claude|gemini|copilot|grok|deepseek|mistral. You immediately see which engines already send you visits, in what proportions, and how GA4 classifies them today.

2. Check what the native channel captures

In your acquisition reports, look for the "AI Assistant" channel (default channel group). Compare its total with your exploration from step 1: the gap corresponds to the engines Google does not yet recognize, which remain classified as Referral.

3. Create a custom "AI Traffic" channel group

In Admin, then Data display, then Channel groups, duplicate the default group and add an "AI Traffic" channel with the condition "Source matches regex":

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

Place this channel above Referral in the evaluation order, otherwise Referral captures the sessions before it. Adapt the list to your observations from step 1, and revalidate it with every new engine: it is a living list.

4. Look at landing pages, not just volume

Cross the AI Traffic channel with the "Landing page" dimension. It is the most useful reading: it shows which content the engines cite with a link that actually gets clicked, and therefore where your GEO work produces visits. On our properties and in our audits, it is almost always a handful of highly extractable pages that concentrates most of this traffic, rarely the homepage.

5. Track the UTMs the engines add themselves

Since June 2025, ChatGPT adds a utm_source parameter to links in its answers that did not carry one yet (announcement reported by PPC Land), and in early October 2026 Google started adding still-undocumented UTMs to Gemini's outbound links (Search Engine Journal). These parameters offer a second attribution anchor when the referrer is missing: watch the Campaign dimension and source values containing "chatgpt" or "gemini", and avoid overwriting them with your own UTM clean-up rules.

What GA4 will never see: read your curves as a floor

The AI traffic you see in GA4 is a floor, not a total. Three structural leaks, confirmed by the documentation and market observation, explain it. The assistants' mobile and desktop apps rarely pass the referrer: those visits fall into Direct. Clicks from AI Overviews and AI Mode remain counted in Organic Search, by design. And above all, most of the AIs' influence produces no click at all: when ChatGPT answers by citing your brand without the user visiting your site, GA4 sees nothing, even though the recommendation happened. It is the most frequent finding of our audits: brands regularly cited in the answers, with near-zero AI referral traffic, wrongly conclude that "AI brings them nothing". Citation is measured by querying the engines themselves, not only in your analytics.

In practice: interpret trends (growth, which pages capture, relative conversion) rather than absolute values, and set these curves against your GEO visibility KPIs (citation rate, share of voice) to connect the cause (being cited) to the effect (being visited).

E-commerce, B2B: two readings of the same channel

For an e-commerce site, the strong signal is quality: with the conversion and revenue-per-visit levels reported by Adobe, even a modest volume of AI sessions deserves its own reports, and extractable product pages become landing pages in their own right. For a B2B or SaaS site, the volume is even smaller but the traffic is ultra-qualified: the question asked to the AI ("best tool for [specific case]") acts as prior qualification. Add a "source" field to your forms or CRM ("How did you hear about us?" captures what GA4 misses, notably visits coming from the apps), and follow the AI Traffic channel through to the demo request, not just to the session.

Frequently asked questions

How do I see ChatGPT traffic in GA4? Three complementary ways: the native "AI Assistant" channel (active since May 2026), an exploration filtered on sources containing chatgpt.com or chat.openai.com, and a custom channel group with a regex covering all the engines. The custom channel is the most complete.

Does GA4 have a dedicated AI traffic channel? Yes. Since May 2026, the default channel group includes "AI Assistant": GA4 assigns the medium "ai-assistant" when the referrer matches its list of recognized assistants (ChatGPT, Gemini, Deepseek, Copilot, Grok cited as examples). The full list is not published and AI Overviews are excluded from it.

Why does my AI traffic look so small? Because a large part is not attributable: the assistants' apps rarely pass the referrer (visits counted as Direct), AI Overviews clicks stay in Organic Search, and most AI recommendations generate no click at all. The GA4 number is a floor; the trend matters more than the volume.

Do clicks from AI Overviews count as AI traffic? No. Google's documentation explicitly excludes AI Overviews and AI Mode from the AI Assistant channel: those clicks remain counted in Organic Search, with no GA4 dimension to isolate them. Search Console does not distinguish them from classic results either.

Which regex should I use to isolate AI traffic? A condition on the session source such as (^|\.)(chatgpt\.com|chat\.openai\.com|perplexity\.ai|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com|grok\.com|deepseek\.com|mistral\.ai|chat\.mistral\.ai)$, placed above Referral in a custom channel group, and updated whenever new engines appear.

Does traffic from AIs convert better? On US retail, yes: according to Adobe, this traffic converted 42% better than other channels in March 2026, with revenue per visit 37% higher, whereas it converted worse a year earlier. Check on your own data: that is precisely what a well-configured AI Traffic channel enables.

Conclusion

Measuring AI traffic in GA4 has become simple: a native channel, a complementary regex, a landing-page reading. The essential part is knowing its limits: what you see is the clicked, attributable share of a much broader influence that plays out inside the answers themselves. The complete measurement therefore has two sides: in GA4, the visits the engines send you; on the engines' side, the citations they grant your brand, question by question. The topic moves fast (the native channel is five months old, Gemini's UTMs a few days), and we will keep this guide updated. For the side GA4 does not show, the free Sapian Metrics audit queries 8 engines, including ChatGPT and Google AI Overviews, and shows you in minutes where your brand is cited, against your competitors.