Measurement

How to track AI referral traffic in GA4, and why every number you get is an undercount

By Arnav Mukherjee, founder of TofuBofu · July 13, 2026

When we built AI traffic attribution into TofuBofu, the first thing we had to accept was uncomfortable: a meaningful share of AI-driven visits cannot be attributed by anyone, not by us, not by GA4, not by the fanciest enterprise analytics suite. The referrer simply is not there. We designed our own cookieless tracker around that honesty, and it changed how I read every AI traffic report I see.

So this post does two jobs. First, the practical setup: exactly how to surface ChatGPT, Perplexity, Gemini, Claude, and Copilot visitors in GA4, which takes about fifteen minutes. Second, the part most guides skip: the three structural leaks that make your AI traffic number a floor rather than a total, and what to measure instead so the leaks do not mislead you.

The 15-minute GA4 setup

When someone clicks a link inside an AI answer on the web, their browser usually passes a referrer, and GA4 records it as the session source. The sources you are looking for: chatgpt.com (and the older chat.openai.com), perplexity.ai, gemini.google.com, copilot.microsoft.com, and claude.ai. GA4's default channel grouping files these under generic Referral, which is why AI traffic hides in plain sight.

Quick look (2 minutes): Reports → Acquisition → Traffic acquisition, switch the primary dimension to Session source / medium, and type "chatgpt" or "perplexity" in the search box. That is your raw feed.

Always-on view (10 minutes): Admin → Data settings → Channel groups → create a custom channel group, add a channel named "AI referrals" where session source matches the regex:

Put the AI channel above Referral in the evaluation order so it wins the match. Then mark your real conversions, form fills, booked calls, signups, as key events, so you can read AI-referred conversion rate and not just sessions. Revisit the regex quarterly: engines add domains, and new engines keep appearing.

The three leaks: where AI traffic hides from you

Leak one: apps strip the referrer. A large share of ChatGPT usage happens in its mobile and desktop apps, and clicks from apps frequently arrive with no referrer at all. GA4 files them under Direct. Same for links a user copies from an answer and pastes into a browser.

Leak two: AI Overviews wears a disguise. A click from a Google AI Overview carries ordinary google / organic attribution. There is no dimension in GA4 or Search Console that separates it from a classic blue-link click. The engine with the largest reach is the one whose AI traffic you structurally cannot isolate.

Leak three: the biggest slice never clicks. AI answers resolve questions in place. A buyer who reads a recommendation, then visits your site tomorrow by typing your name, registers as Direct or branded search. The influence was AI; the analytics say otherwise. This is not a defect you can configure away, it is what answer-first discovery does to click-based measurement, the same shift I wrote about in will AI search kill my SEO traffic.

Measure the answers, not just the clicks

A free TofuBofu scan shows whether AI engines recommend you at all, the leading indicator that analytics can only confirm after the fact.

Run your free scan

Reading the numbers without fooling yourself

1. Treat every AI traffic figure as a floor

The labeled sessions are the slice that survived all three leaks. Trend it, do not total it. If labeled AI sessions doubled, the real influence probably grew at least as fast; if they are flat at near-zero, check your presence in the answers before concluding the channel is dead for you.

2. Judge the channel on conversion, not volume

AI-referred visitors arrive pre-qualified: the engine already matched your firm against their situation before they clicked. Small session counts with strong conversion is exactly what a young, high-intent channel looks like. Mark key events and compare conversion rates across channels.

3. Watch branded search and Direct alongside it

Because leak three routes AI influence into those buckets, a rise in branded queries and Direct visits that tracks your AI visibility improvements is part of the same story. Read the three lines together.

4. Pair analytics with answer scanning

Analytics tells you what happened after a click. Scanning the answers tells you whether you are in them at all, which is upstream of everything and immune to referrer loss. Presence is the leading indicator; traffic is the lagging confirmation.

One note on how we handle this in our own product, because it shaped the advice above. TofuBofu's built-in attribution is deliberately cookieless: it tags AI-referred visits and conversions in aggregate, on your own domain, with no cross-site cookies and no visitor identity. We accepted the undercount in exchange for measurement we can defend. Whatever tool you use, insist on knowing which of the three leaks it has actually solved, because vendors claiming complete AI attribution have usually just relabeled the same referrer data.

Frequently asked questions

How do I see ChatGPT traffic in GA4?

Visitors who click a link in a ChatGPT answer arrive with the referrer chatgpt.com, which GA4 files under session source 'chatgpt.com' in the Referral channel. To see it: open Reports, then Traffic acquisition, change the primary dimension to Session source / medium, and search for chatgpt. For an always-on view, build a custom channel group or an Exploration filtered on a regex covering all AI referrer domains.

What regex matches AI referral traffic in GA4?

A solid starting pattern for session source is: chatgpt.com|chat.openai.com|perplexity.ai|gemini.google.com|copilot.microsoft.com|claude.ai|.openai.com. Use it in an Exploration filter (session source matches regex) or as the condition for a custom 'AI referrals' channel in a custom channel group. Revisit it quarterly: engines add domains, and new engines appear.

Why does my AI traffic number look so small?

Partly because the channel is genuinely young, and partly because GA4 structurally undercounts it. Visits from AI mobile apps and desktop apps often carry no referrer and land in Direct. Clicks from Google AI Overviews arrive as google / organic, indistinguishable from a normal search click. And many AI-influenced buyers never click at all: they read the answer, then later type your name in directly, which shows up as Direct or branded search. Whatever GA4 shows you labeled as AI is a floor, not a total.

Can I track traffic from Google AI Overviews separately?

Not in GA4. Clicks from AI Overviews carry the same google / organic attribution as ordinary search results, and Google does not expose a separate referrer or dimension for them, and Search Console does not break them out either. The practical approach is to track your presence in AI Overviews directly (whether you are cited for your queries) rather than waiting for analytics to reveal it.

Is AI referral traffic worth tracking if the volume is small?

Yes, for two reasons. First, quality: AI-referred visitors arrive pre-qualified, because the engine already matched them to you against their situation, so conversion behavior matters more than session counts. Second, trend: the channel is growing from a small base, and the firms that instrument it now will know when it inflects for their category. Mark your key conversions and watch AI-referred conversion rate, not just sessions.

Should I measure AI visibility through analytics or through the answers themselves?

Both, but know what each measures. Analytics measures the click-through slice of AI influence, and misses readers who never click. Scanning the answers themselves measures whether you are present and recommended at all, which is upstream of every click and immune to the referrer leaks. Treat GA4 as the outcome check and answer-scanning as the leading indicator.

Sources and further reading

Google Analytics: custom channel groups: the official documentation for the channel-group setup above
Orbit Media: tracking AI traffic and leads in GA4: a thorough independent walkthrough with report screenshots
Swydo: the agency guide to AI traffic in GA4: alternative regex patterns and agency reporting angles
G2 2026 AI Search Insight Report: the buyer-behavior data behind why this channel deserves instrumentation now