By 2026, AI engines like ChatGPT, Perplexity and Gemini send a measurable and growing share of referral visits — yet in a default GA4 property almost none of it is labeled as such. The traffic is real; the reporting is blind. Most of it lands in the 'direct' bucket or blends into an undifferentiated Organic channel, so teams that are clearly being cited by AI cannot prove a single session came from one.
This guide builds the GA4 channel rules, regex segments and UTM strategy that surface ChatGPT, Perplexity, Gemini and Copilot traffic — and it is deliberately honest about the blind spots, like AI Overviews, that no segment can fully close. To check how AI-ready and well-measured your content is today, run our free 5-axis content audit.
Updated 2026-05-25 with the current AI referrer hostnames, GA4 channel behavior and measurement limits observed across US, UK and European properties.
- AI traffic hides in the direct bucket and an undifferentiated Organic channel — default reports miss it. 2. Build a custom channel group with a referrer regex over chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com. 3. AI Overviews are the hard blind spot — those clicks still read as google.com and cannot be cleanly isolated. 4. UTM-tag every link you control to reclaim attributable AI sessions from direct. 5. Report a floor, not a total — disclose the app and AI Overview blind spots every time.
Why is AI and LLM traffic invisible in default GA4 reports?
Default GA4 was built for a web of links, and AI surfaces break two of its core assumptions. The result is that genuine AI-driven visits are scattered across the wrong channels.
The direct bucket — GA4 files a session as direct whenever it cannot determine a source. App-based assistants — the ChatGPT app, a Copilot panel, a mobile client — frequently send no referrer header, so their clicks land in direct, indistinguishable from someone typing your URL.
The organic blur — A click from a Google AI Overview is still a click from google.com, so it reads as ordinary Organic Search. There is no distinct hostname to separate it, which means a growing slice of AI-influenced traffic hides inside a channel you already have.
No default AI channel — GA4 ships with Organic, Paid, Referral and Direct, but nothing for AI engines, so even referral-bearing AI clicks get filed under generic Referral. The fix starts with naming the surface yourself. For the strategy that earns this traffic in the first place, see our complete GEO guide.
Which referrers identify ChatGPT, Perplexity, Gemini and Copilot?
You cannot segment what you cannot name. The first step is a maintained list of the hostnames each engine passes when it does send a referrer.
ChatGPT — primarily chatgpt.com, plus the legacy chat.openai.com. Web sessions usually carry one of these; app sessions often carry none.
Perplexity — perplexity.ai. Perplexity is one of the more reliable referrers because it frequently links out with attribution intact.
Gemini and Copilot — gemini.google.com for Gemini, and copilot.microsoft.com plus bing.com for Microsoft Copilot. Add claude.ai, you.com and poe.com to round out the major surfaces.
Keep all of these in a single regular expression so the list is easy to maintain, and review it every quarter — new engines launch and hostnames change. For the upstream work of actually earning these citations, our guide to being cited by AI engines covers the content side.
How do you build a custom channel group and regex segment?
GA4 gives you two complementary tools: a custom channel group for ongoing reporting and an exploration segment for analysis. Build both.
Custom channel group — In Admin, create a custom channel group based on the default, then add a channel named AI / LLM. Define its condition as the session source matching a regex of your AI hostnames, and order it above the generic Referral channel so AI traffic is not miscredited.
The regex — Join your hostnames with the OR operator, for example chatgpt\.com|chat\.openai\.com|perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com|claude\.ai|you\.com. Escape the dots so the pattern is precise.
Exploration segment — In Explore, create a session segment on the same condition to analyze landing pages, engagement and conversions for AI visits. This shows behavior, not just volume. When the numbers look strange, our GA4 explorations guide goes deeper on segment building.
How do AI Overviews traffic and 'direct' inflation behave?
These are the two failure modes that no regex fully solves, and pretending otherwise produces a report that looks complete but is not.
AI Overviews — A click from an AI Overview is a click from google.com, so GA4 records the same organic referrer it always has. There is no ai-overview hostname to match, which means this traffic blends into Organic Search and cannot be isolated by referrer alone. You can infer movement from Search Console landing-page and query trends, but you cannot draw a hard boundary in GA4.
Direct inflation — As app-based assistants strip referrers, the share of genuine AI visits landing in direct grows. A rising direct channel with no obvious cause is often a sign of AI activity you cannot see, especially if it correlates with content the engines are known to cite.
The honest read — Treat both as structural undercounts. The captured segment is a floor; the real influence is larger. If a traffic shift looks like a tracking artifact rather than a performance problem, the same separation discipline in our conversion-tracking fix guide applies here.
How do you use UTMs where you control the link?
You cannot tag a link an AI engine writes on its own, but you can tag every link you place in content those engines read. That is the slice you can reclaim from direct.
Tag what you own — Apply UTMs to URLs in your documentation, knowledge base, llms.txt, social profiles and syndicated posts. When an engine surfaces one of those links, the click arrives attributed instead of anonymous.
Keep the scheme consistent — Use a stable utm_source and utm_medium convention so AI placements roll up cleanly. Inconsistent tagging fragments the data worse than no tagging at all.
Connect it to exposure — The more of your content is machine-readable, the more controlled links exist to tag. Publishing an llms.txt index, covered in our llms.txt implementation guide, both exposes content and creates taggable surfaces. Build links fast with the UTM builder.
What are the measurement blind spots to disclose?
A trustworthy AI-traffic report names its own limits. Hiding the gaps makes the number look authoritative and quietly wrong.
App clicks — Sessions from assistant apps often carry no referrer and fall into direct. You will undercount every engine with a popular app, and you cannot fully recover those visits.
AI Overviews — As covered above, these read as organic google.com and cannot be cleanly separated. Report organic movement and Search Console signals instead of a false-precision AI Overview number.
Sampling and stripping — Some referrers are dropped by privacy settings or sampled out of high-volume reports. Cross-check with server logs, which see raw referrers GA4 may not surface.
The reporting frame — Present a captured minimum plus a labeled blind-spot list, never a single total dressed up as complete. This is the same intellectual honesty our AI Overviews impact guide applies to performance claims.
How AI traffic tracking fits a broader GEO strategy
Measurement is the feedback loop for Generative Engine Optimization. You cannot improve what you refuse to estimate, and you cannot trust an estimate that hides its gaps.
Measure to steer — A maintained AI channel tells you which content earns citations and clicks, so you can do more of what works. Without it, GEO is guesswork.
Pair earning with measuring — Earning citations and measuring the return are two halves of the same loop. The content side lives in our guide to ranking in AI Overviews; the measurement side is this article.
Close the loop — Audit content quality and AI-readiness, ship the channel and segment, tag your controlled links, and report a floor with honest blind spots. To bring content, structure and measurement together, run the SteerAds free 5-axis audit and tag every controlled link with the UTM builder.
Sources
Official and primary sources consulted for this guide:
-
support.google.com — GA4 default channel groups
-
platform.openai.com — OpenAI bots and crawlers
-
llmstxt.org — the llms.txt proposal
-
blog.google — Google Search and AI Overviews
FAQ
How do I track ChatGPT traffic in GA4?
Build a referral-based segment or custom channel group that matches the hostnames ChatGPT uses, primarily chatgpt.com and the older chat.openai.com. In GA4, create a custom channel group, add a channel named something like AI / LLM, and define its condition as the session source matching a regex of those hostnames. You can also build an exploration with a segment on the same condition to analyze behavior. The catch is that not every ChatGPT click carries a referrer — clicks from the desktop and mobile apps often arrive without one, so they fall into direct and never reach your segment. Treat the number you capture as a floor, not a complete count.
Does GA4 show AI Overviews traffic?
Mostly no, and this is the hardest blind spot. A click from a Google AI Overview is still a click from google.com, so GA4 sees the same organic search referrer it always has — there is no distinct ai-overview hostname to segment on. That means AI Overview traffic blends into your existing Organic Search channel and cannot be cleanly isolated by referrer alone. You can infer movement by watching landing-page and query-level shifts in Search Console alongside organic sessions, but you cannot draw a hard line around AI Overview clicks in GA4 today. Be explicit about that limitation when you report.
Why is AI traffic counted as direct in GA4?
Because the referrer is missing. GA4 files a session as direct whenever no source can be determined, and AI surfaces strip or omit the referrer more often than classic websites. App-based assistants — the ChatGPT app, a Copilot panel, a mobile client — frequently send no referrer header at all, and some engines deliberately do not pass one. The result is that a meaningful share of genuine AI-driven visits lands in direct and is indistinguishable from someone typing your URL. UTM-tagging the links you control is the only reliable way to pull some of that traffic back out of the direct bucket.
Can I build an AI channel in GA4?
Yes. GA4 lets you create a custom channel group, where you add a new channel — call it AI / LLM — and define its matching rule with a regular expression over the session source or referrer. List the hostnames you care about, such as chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com, joined with the regex OR operator. Custom channel groups apply going forward and to some historical data depending on your setup, and they keep AI traffic from being miscredited to Organic or Referral. It is the cleanest single step you can take, provided you accept that it only captures sessions that actually carry one of those referrers.
How accurate is AI traffic tracking in GA4?
Partial, and you should say so out loud. What you capture is a reliable floor — sessions that genuinely carried an AI referrer — but the true number is higher because app clicks, stripped referrers and AI Overviews leak into direct and organic. Expect to undercount, sometimes substantially, and never present the segment as a complete picture. The right framing in a report is a captured minimum plus a clearly labeled set of blind spots. Pair the GA4 segment with server-log analysis and Search Console trends to triangulate, and update your referrer regex as new engines appear, because the list changes every quarter.
Which referrers should I include for AI engines?
Start with the major surfaces and expand as you see traffic. The core list in 2026 is chatgpt.com and chat.openai.com for ChatGPT, perplexity.ai for Perplexity, gemini.google.com for Gemini, copilot.microsoft.com and bing.com for Microsoft Copilot, and claude.ai for Claude. Add you.com, poe.com and any vertical assistant relevant to your market. Keep the list in one regex so it is easy to maintain, and review it quarterly because new engines launch and hostnames change. Remember that this captures referral-bearing sessions only; the app and AI Overview blind spots remain regardless of how complete your hostname list is.
Should I use UTMs for AI traffic?
Yes, wherever you control the link. You cannot add UTMs to a link an AI engine generates on its own, but you can tag every URL you place in content the engines read — your docs, your knowledge base, your llms.txt, your profiles and your syndicated posts. A consistent UTM scheme turns those controlled placements from anonymous direct hits into attributable sessions. It will never capture organic AI citations, but it reclaims the slice you can influence, and it is the single highest-leverage habit for measuring AI referral traffic over time.