How to track AI traffic in Google Analytics 4 (and why it undercounts)

One day you open Google Analytics, filter by source, and there's a new row: chatgpt.com / referral, with a handful of sessions. The question is unavoidable: how many people actually reach my site from AI? GA4 can answer part of it —and since May 2026 it does so on its own, with no setup— but that number is the tip of the iceberg. It's worth understanding exactly what it shows you and, above all, what slips past it.

Let's go in order: first the quick way to see it, then how to capture what the automatic mode doesn't recognize, and finally what actually matters —why that number always undercounts and which question it doesn't answer.

The fastest way: see it without setting anything up

Before touching any setting, do the thirty-second check. In GA4, go to Reports → Acquisition → Traffic acquisition, switch the dimension dropdown to Session source / medium, and type "chatgpt" into the table's search box. If someone arrived from there with a visible referrer, you'll see rows like chatgpt.com / referral. Repeat with perplexity, gemini, claude or copilot and you'll have a first map of who's sending you people.

It's crude, but it confirms the channel exists before you invest time building anything. If nothing shows up there at all, it doesn't mean the measurement is broken: you may genuinely not be getting AI clicks with an identifiable referrer yet. Below you'll see why that's more common than it looks.

Until May 2026 it landed in "Referral." Now there's a native channel

For a long time, all chatbot traffic landed mixed inside the generic "Referral" bucket, alongside every other site that linked to you. In May 2026, Google changed that: it added a native channel called "AI Assistant" to GA4's Default Channel Group. When Analytics detects a referrer it recognizes as an AI assistant, it assigns the ai-assistant medium and groups those sessions into their own channel —no filters, no regex, not a single expression to write.

The fine print matters. Google names ChatGPT, Gemini and Claude as examples of recognized assistants, but it hasn't published the full list, and at launch it didn't confirm that Perplexity or Copilot were part of that automatic recognition. The rollout is also gradual: it may take several weeks to reach 100% of properties, so if you don't see the channel yet, it isn't your mistake. If your AI visibility runs through Perplexity or Microsoft Copilot, the native channel alone won't cover you.

To capture Perplexity, Copilot and the rest: a custom channel group

This is where control comes back to you. A custom channel group lets you define exactly which sources count as AI, independent of what Google decides to recognize. Go to Admin → Data display → Channel groups, create a new one, add a channel and name it "AI Traffic." Set the first condition to Session source, the match type to matches regex, and paste a pattern with the domains you care about.

A base pattern

A reasonable starting point, covering the assistants with real traffic today:

chatgpt\.com|chat\.openai\.com|perplexity\.ai|gemini\.google\.com|claude\.ai|copilot\.microsoft\.com|bing\.com/chat|grok\.com|x\.ai|meta\.ai|deepseek\.com

Then drag that channel above "Referral" and "Organic Search" in the list, because GA4 assigns each session to the first channel it matches, top to bottom: if "Referral" sits higher, it grabs your AI traffic before your rule ever sees it. Save, and most acquisition reports will start showing "AI Traffic" as its own channel. Keep the pattern alive: every few months a new assistant shows up with volume, and the regex has to be updated by hand.

The catch: GA4 undercounts, and by a lot

Here's the point almost no tutorial tells you, too busy selling the regex as if it solved the problem. For GA4 to add a click to your AI channel, that click has to arrive with an identifiable referrer. And a huge share of AI traffic doesn't carry one. The assistants' mobile apps, in-app browsers and much of the free sessions don't send the source data: the click happened, but GA4 can't see where it came from and files it as "Direct." Your AI channel reads zero while people keep walking in from a ChatGPT answer.

There's a second leak, even more invisible. Plenty of people see your brand in an AI answer, get no link to click, and end up searching for you on Google by name. That visit goes entirely to "Organic Search": AI generated the demand, but the search engine takes the credit. And the biggest leak of all doesn't even produce a visit —the link-less mention, the recommendation the user reads and postpones— never appears in any GA4 report, no matter how many channels you create.

The takeaway isn't that GA4 is useless. It's useful, and the AI channel is a real improvement. The takeaway is that the number you see is a floor, not a portrait: real AI traffic is always larger than the label. Treating that figure as "all of my AI impact" is the expensive mistake.

Traffic isn't recommendation: two different questions

It's worth separating the two questions people conflate all the time. One is how much traffic AI sends me: that's what GA4 measures, with its limits. The other is whether AI recommends me: whether, when someone asks ChatGPT or Perplexity for a tool or a provider in your category, your name shows up in the answer, in what position and with what description.

The second one doesn't show up in analytics, because most of those conversations never end in a click to your domain. It's measured by querying the models directly with your sector's questions and reading what they answer —which is exactly what we cover in the guide on how to measure if ChatGPT recommends you. The two metrics complement each other: GA4 gives you the trickle of clicks that do arrive; mention testing gives you the picture of your reputation inside the model, which is what drives that trickle.

What to do this week

If you had to do three things, they'd be these. First, open GA4 and check whether you already have the "AI Assistant" channel; if you do, look at the last few weeks' volume to set a baseline. Second, build the custom channel group with the regex above, so you don't depend only on the three assistants Google recognizes and you capture Perplexity and Copilot. Third, and most important, don't mistake that number for your AI visibility: put a mention measurement next to it, because that's the one that tells you whether AI recommends you before the click ever exists.

With that trio done, you'll have both halves of the picture —the traffic that arrives and the recommendation that originates it— instead of a single figure that, without context, would have you underestimating exactly the channel that's growing fastest.

Frequently asked questions