Two numbers, never summed

How do you see AI traffic in your analytics - and what will you never see?

You can build a channel group that catches every click an AI answer sends you. It won't catch the recommendation that never got clicked - and that's most of them.

28 Labs · August 2026

Segment your referrer data for chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com, and claude.ai, and use GA4's native AI Assistant channel (added May 13, 2026) plus a custom channel group for the surfaces it doesn't cover yet. That shows you AI-sourced traffic - visitors whose first touch was a click from an AI answer. It will never show you AI-influenced traffic - buyers an AI answer talked into your brand who then arrived by typing your name into Google or opening your app directly. Those are two different numbers, built from two different instruments, and you should never add them together.

How do you set up GA4 to catch AI referral traffic?

1
Check the native AI Assistant channel
2
Add a custom channel group for the gaps
3
Reorder rules above Referral
4
Save an exploration segmented by source
Ten minutes of setup. None of it is exotic - it's the same channel-group mechanics GA4 has used for years, pointed at a new list of domains.

Google added a native AI Assistant channel to GA4's Default Channel Group on May 13, 2026. Any session where the referrer matches a recognized AI domain now gets tagged ai-assistant automatically - no configuration required. Google names ChatGPT, Gemini, and Claude as the covered sources.

It doesn't cover everything. Perplexity and Microsoft Copilot aren't in the native list, so their traffic still falls into the plain Referral bucket unless you build your own channel group with a regex across chatgpt.com, chat.openai.com, perplexity.ai, gemini.google.com, copilot.microsoft.com, and claude.ai. Put that custom rule above your Referral rule - GA4 evaluates channel definitions top to bottom, and if Referral fires first, your AI traffic quietly disappears into it. Then save an exploration that layers source/medium against conversions, so the number isn't buried in an acquisition report nobody opens twice.

What does utm_source=chatgpt.com actually tell you?

ChatGPT tags a share of its outbound links with utm_source=chatgpt.com. OpenAI broadened this in mid-2025 to cover more of ChatGPT's search and browsing links, and it's now common enough that most sites with any AI-driven traffic will see it in their Traffic Acquisition report under the chatgpt.com referral source. When you see it, it's a genuine signal: a user asked ChatGPT something, ChatGPT pointed at your page, and the user clicked.

The problem is what it doesn't tell you: whether the click that didn't carry the tag also came from ChatGPT. Referrer headers get stripped inside ChatGPT's and Perplexity's native mobile apps, which open links in embedded browsers that don't always pass a referrer at all. Some outbound links carry rel="noreferrer". Copy-pasted links carry nothing. All of that traffic lands in GA4 as Direct, indistinguishable from someone who typed your URL from memory. The parameter you can see is real. It's a floor, not a ceiling.

What does the traffic actually look like once you can see it?

Adobe's Digital Insights team, in its Q2 2026 AI traffic report published in April, found AI-driven visits to retail sites grew 393% year over year in the first three months of 2026, and that by March 2026 those visitors converted 42% better than non-AI traffic - a reversal from a year earlier, when AI-referred visitors were converting at roughly half the rate of everyone else. Adobe also found AI-referred sessions spent more time on-site and viewed more pages per visit than the non-AI baseline.

Two things follow. First, the volume most sites see through referrer data alone is still small relative to organic and paid - this is a channel in its early growth curve, not a mature one. Second, small and high-intent isn't a contradiction. A visitor who arrives already having had an AI system explain your product to them converts differently than one still comparing ten blue links. That's exactly why the click you can measure is worth more attention than its volume suggests - and exactly why it still isn't the whole picture.

AI-sourced vs AI-influenced: the two numbers you can't sum

AI-sourced (you can see this)

  • First touch was a click from an AI answer
  • Measured by referrer domain or UTM tag
  • Lives in GA4, in your acquisition reports
  • Always the smaller, floor number

AI-influenced (you can't)

  • An AI answer shaped the decision, at any point
  • Arrived via direct, brand search, or a marketplace
  • Invisible to referrer data entirely
  • Only found by asking the buyer directly
Different instruments, different denominators. Report them side by side. Never add them into one "AI traffic" figure.

This is the distinction that matters more than any setup step above it. AI-sourced is what analytics measures: the click that carried evidence of where it came from. AI-influenced is everyone an AI answer moved toward you who then converted through a channel analytics attributes to something else entirely - a Google brand search, a direct visit, a listing on a marketplace. We cover the mechanics of what visibility tooling can and can't observe in what AI visibility tools can and can't measure, and the answer-layer thinking behind it in what is AI discovery intelligence.

Most exposure never produces a clickAn AI engine can recommend your brand by name to a user who never clicks anything - they read the answer, close the tab, and walk into your store or type your name into Google a week later. Referrer analytics has no instrument for that moment. It only ever measures the minority of AI exposure that ends in an immediate, trackable click. Report your GA4 AI numbers honestly as "the traffic we can see," not "the AI channel."

The cheapest attribution instrument you're not using

You can't fix analytics' blind spot with more analytics. You fix it at the point of sale. Add one mandatory field to your checkout, booking form, or signup flow: "How did you hear about us?", with an explicit option like "ChatGPT / AI tool" alongside Google, referral, and social. It costs nothing to build, takes one dropdown, and it's the only instrument that catches the buyer who read your answer in Perplexity on Tuesday and bought on Google Search on Friday.

Run it long enough and you get a self-reported influence rate to sit next to your referrer-measured AI-sourced rate. They won't match. They're not supposed to. The gap between them is the size of the exposure your analytics has been missing the whole time.

What we do about it

We don't rely on referrer data to tell us whether AI engines recommend a brand. We ask ChatGPT, Claude, Gemini, and Perplexity the real questions a buyer would ask, on a fixed schedule, and read what each engine actually says. That measures the answer itself - visible or not, clicked or not - instead of the fraction of it that happened to leave a UTM tag behind. Referrer-based AI traffic is a useful, honest, incomplete number. We build the instrument for the number GA4 was never built to see.

28 Labs measures how AI engines answer real buyer questions - and what moves the counts. Every number we publish re-derives to specific questions and engines. try28labs.com