Kubnal Bridge
Back to Blog

How to Measure AI Visibility and Track AI Referral Traffic in GA4

Kubnal Bridge Editorial TeamJuly 28, 20268 min readGEO / AEO
Title card reading How to Measure AI Visibility in GA4, from kubnalbridge.com

AI referral traffic grew 66% in 2025, yet most of it never shows up as AI in your analytics. In 2026, Semrush's traffic channel study found AI drives roughly 0.14% of all visits, while Direct sits at 64.69% (Semrush, 2026).

So a large slice of your AI traffic is hiding inside the Direct bucket. You can't prove AI sends visits, and you can't tell whether the engines name you at all. This guide gives you a two-layer method. First, build a GA4 channel group that isolates AI referrals. Second, run share-of-voice testing for the pre-click visibility GA4 can never capture. We'll cover the native channel Google added, its Perplexity blind spot, and what still stays invisible. If the wider strategy is new to you, start with our pillar on generative engine optimization.

Key Takeaways

  • Much AI traffic lands in GA4 as "Direct" because app-based AI visits arrive with no referrer; a custom channel group recovers the web portion.
  • GA4 added a native "AI Assistants" channel in 2026, but Google's list excludes Perplexity, which still lands in Referral.
  • AI referral traffic is tiny but exploding: about 0.14% of visits, up 66% in 2025 (Semrush, 2026).
  • It converts well: ChatGPT referrals convert at 7.1%, second only to paid search (Similarweb, 2026).

Why Does ChatGPT Traffic Show Up as "Direct" in GA4?

Because a large share of AI visits arrive with no referrer. Native iOS and Android AI apps open links without passing a source, so GA4 files those sessions under Direct. In 2026, Terminus documented this referrerless behavior as the main reason ChatGPT traffic hides in the Direct bucket (Terminus, 2026).

Here's the split worth understanding. A desktop web visit from chatgpt.com usually carries a referrer, so GA4 can attribute it. An in-app tap on the same link often carries nothing. GA4 sees a session with no source and no medium, and its only honest option is Direct.

That matters because Direct is already your noisiest channel. In 2026, Semrush measured Direct at 64.69% of all visits, with Organic at 16.04% and AI at roughly 0.14%, up 66% year over year (Semrush, 2026). When AI referrals pour into a bucket that big, they vanish into the noise. You never see the growth, and you can't defend the channel in a report.

This blind spot is one reason many teams underestimate the shift. We unpack the wider erosion pattern in is AI search eroding your organic traffic.

Does GA4 Now Track AI Traffic Automatically?

Partly. In 2026, GA4 added a native "AI Assistants" default channel that, in Google's own documentation, recognizes ChatGPT, Gemini, DeepSeek, Copilot, and Grok (Google Analytics Help, 2026). It's a genuine step forward. But Google's list does not include Perplexity, which still lands in Referral, so a custom group still earns its keep.

Treat the native channel as a helpful default, not a complete solution. It captures the biggest engines automatically, which spares you some regex. What it won't do is guarantee coverage of every source your buyers use, and the gaps move as engines launch and rename.

When we set this up for clients, the second surprise is always the same: Perplexity isn't in Google's native channel. Google's own default-channel documentation lists ChatGPT, Gemini, DeepSeek, Copilot, and Grok, and Perplexity is not among them. So we capture it manually rather than assume coverage the docs don't confirm.

The practical takeaway is simple. Keep the native channel on, then build a custom group on top of it. That way you own the definition, and a future engine launch is a one-line edit rather than a blind spot you discover months late.

How Do You Build a GA4 Channel Group for AI Traffic?

In Admin, open Data display, then Channel groups, and create a custom channel named "AI Assistant." Add a "Session source matches regex" condition, then move it above Referral in evaluation order. In 2026, Terminus laid out this exact sequence as the reliable way to isolate web AI referrals (Terminus, 2026).

The evaluation order is the step most guides skip. GA4 assigns each session to the first matching channel, top down. If Referral sits above your new channel, it grabs the AI hostnames first, and your custom channel stays empty. Drag "AI Assistant" above Referral, and the matches flow where you want them.

Here's a copy-ready regex covering the confirmed AI hostnames:

`` chatgpt\.com|chat\.openai\.com|perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com|claude\.ai|deepseek\.com|x\.ai|grok\.com|meta\.ai|you\.com|poe\.com ``

Paste that into the "Session source matches regex" field. Add or trim hostnames as engines change. Keep chatgpt.com and chat.openai.com both, since either can appear.

The first thing teams notice after the change is how much traffic reclassifies out of Direct. Once the regex goes live, a slice of what looked like "Direct" suddenly reads as AI Assistant, and the trend line finally becomes defensible. We don't quote a fixed percentage, because it depends on how app-heavy your audience is, but the reclassification is always visible.

Why bother at all? Because this is not vanity traffic. The chart below shows sign-up conversion by channel, and the LLM line sits well above search and direct.

Sign-up conversion rate by channel, 2025 Sign-up conversion rate by channel LLM 1.66% Social 0.46% Search 0.15% Direct 0.13% Source: Digiday, 2025
Source: Digiday, 2025.

In 2025, Digiday reported LLM sign-up conversion at 1.66%, against 0.46% for social, 0.15% for search, and 0.13% for direct (Digiday, 2025). That gap is the business case for setting the channel up in the first place.

Does AI Referral Traffic Actually Convert?

Yes, better than almost every other channel. In 2026, Similarweb found ChatGPT referrals convert at 7.1%, second only to paid search at 7.8%, and those visitors view about 24% more pages per visit (Similarweb, 2026). So the traffic is small in volume but high in intent, which is exactly what you want to isolate.

The reason is intuitive. Someone arriving from an AI answer has usually already read a recommendation. They land warmer than a cold search click, closer to a decision. That's why the conversion numbers keep landing near the top of every channel study, from two independent measurement sources.

There's a catch worth stating plainly: almost all of this referral volume comes from one engine. In 2025, Conductor found ChatGPT accounts for 87.4% of AI referral traffic (Conductor, via Digiday, 2025).

ChatGPT share of AI referral traffic, 2025 ChatGPT share of AI referral traffic 87.4% ChatGPT ChatGPT, 87.4% All other AI engines, 12.6% Source: Conductor, 2025. Previsible: 92.4% across 6.77M sessions.
Source: Conductor, 2025. Previsible separately found ChatGPT drove 92.4% of AI referral traffic across 6.77 million sessions ([Search Engine Land](https://searchengineland.com/chatgpt-ai-referral-traffic-sessions-data-481630), 2026).

That concentration cuts two ways. It makes ChatGPT your priority for capture and testing today. It also means a single engine's product change could reshape your numbers overnight, so don't build a strategy on one source alone.

Measure AI Visibility Beyond What GA4 Can See

GA4 only measures AI traffic after a click. A referrerless share stays in Direct permanently, and pre-click visibility, whether an engine even names you, stays invisible. In 2026, Semrush pegged all AI referral at roughly 0.14% of visits, a figure that understates your true exposure in AI answers (Semrush, 2026).

Here's the distinction that reframes the whole problem. GA4 answers "did someone click through from an AI answer?" It cannot answer "am I even in the answer?" Those are two different questions, and most AI exposure lives in the second one. Google AI Overviews, for instance, often resolves the query in place, so there's frequently no referral click to record at all. That's different from a ChatGPT chat that links out.

So you need a second measurement layer. Run your target queries across ChatGPT, Perplexity, and Google AI Overviews on a set schedule, and log who gets cited. Score each engine on its own, because the three name almost entirely different sources. This is share-of-voice testing, and it's the only way to see the pre-click visibility GA4 skips. We walk through the citation side in how to get cited by ChatGPT, Perplexity, and Google AI Overviews.

Running both layers by hand is a real time cost. Agencies such as Kubnal Bridge build this measurement layer when a team can't staff it, pairing the GA4 channel with monthly share-of-voice testing. If you want the evidence base behind the approach, see does GEO actually work.

AI Referral Traffic Is Small but Growing Fast

The direction is steep. In 2026, Semrush reported ChatGPT outbound referrals grew 206% across 2025, approaching a billion monthly visits (Semrush, 2026). In 2025, Similarweb data showed Gemini referrals rose 388% in a single quarter (Similarweb, via Digiday, 2025). Small base, fast curve.

The base numbers back that up. In 2026, Similarweb found total AI referral visits more than tripled year over year, from September 2024 to September 2025 (Similarweb, 2026). A channel tripling annually deserves a measurement setup now, before it matters to your board.

One honest caveat on the forecasts. In 2024, Gartner projected that traditional search volume would fall 25% by 2026 (Gartner, via Search Engine Land, 2024). That drop largely did not materialize, because Google absorbed AI into AI Overviews rather than ceding queries to chatbots. So the shift is real, but it's a reallocation inside search as much as an exodus from it. For the strategic contrast, see GEO vs SEO.

Frequently Asked Questions

Why does ChatGPT traffic show as direct in GA4?

In 2026, Terminus explained that native AI apps open links without passing a referrer, so GA4 defaults those sessions to Direct (Terminus, 2026). Desktop web visits from chatgpt.com are usually attributed, but app traffic hides in the Direct bucket unless you add a custom channel.

Does GA4 track AI traffic automatically now?

Partly. In 2026, GA4 added a native AI Assistants channel covering ChatGPT, Gemini, DeepSeek, Copilot, and Grok (Google Analytics Help, 2026). Google's list excludes Perplexity, which still lands in Referral, so a custom channel group remains necessary for full coverage.

Which sources should I add to my AI channel group?

Include the confirmed AI hostnames: chatgpt.com, chat.openai.com, perplexity.ai, gemini.google.com, and copilot.microsoft.com, plus claude.ai, deepseek.com, x.ai, meta.ai, you.com, and poe.com. In 2026, Terminus recommended placing this regex channel above Referral in evaluation order so the matches land correctly (Terminus, 2026).

Does AI referral traffic convert?

Yes, strongly. In 2026, Similarweb found ChatGPT referrals convert at 7.1%, second only to paid search at 7.8% (Similarweb, 2026). In 2025, Digiday reported LLM sign-up conversion at 1.66%, far above search's 0.15% (Digiday, 2025).

Conclusion

Measuring AI is a two-layer job, and GA4 handles only one of them. Set up the channel group today, then start monthly share-of-voice testing.

  • Build a custom "AI Assistant" channel with a source regex, placed above Referral, to recover web AI referrals from Direct.
  • Keep GA4's native AI Assistants channel on, but capture Perplexity manually, since Google's list excludes it.
  • Value the traffic accordingly: ChatGPT referrals convert at 7.1%, near the top of every channel.
  • Add a second layer, share-of-voice testing, for the pre-click visibility GA4 can never see.

Not sure whether AI engines name you at all? Start with the visibility layer, then wire up the GA4 channel to catch the clicks that follow. For the full picture, return to our pillar on generative engine optimization.

About the Author

Kubnal Bridge Editorial Team. The Kubnal Bridge editorial team researches and writes about generative engine optimization and AI search visibility, drawing on first-hand citation testing across ChatGPT, Perplexity, and Google AI Overviews.

Sources

Back to all articles