How to track AI traffic in GA4

If you want to track AI traffic in GA4, the default reports will fight you: most ChatGPT, Perplexity, and Claude visits land in Direct or get lumped into a generic Referral bucket, so the sessions that answer engines send you stay invisible next to organic search and paid. Google closed part of that gap when it added a native AI Assistant channel to the default channel group on May 13, 2026, reported by Search Engine Journal, but the native channel names only ChatGPT, Gemini, and Claude and still drops any visit that arrives without a referrer. This guide walks through the custom channel group, the referrer regex, and the reporting views that give you a defensible AI traffic number, and it names the one thing most tutorials skip.

Why does AI traffic show up as Direct or Referral in GA4?

Because GA4 classifies sessions by the referrer header, and AI assistants are inconsistent about sending one.

When someone clicks a link inside a web-based chat, the browser usually passes a referrer like chatgpt.com or perplexity.ai. GA4 sees an unknown referring domain, finds no matching rule in the default channel grouping, and files the session under Referral. When someone clicks from a native mobile app, copies and pastes a URL, or uses an in-app browser that strips the referrer, GA4 receives nothing at all and the session falls into Direct. That second bucket is the bigger problem, because a large share of AI usage now happens inside apps rather than desktop browser tabs.

The scale of what you are trying to measure is real. According to Similarweb research, ChatGPT web visits grew 84% between September 2024 and March 2026, Claude grew roughly 770% over the same window, and total AI referral visits across the web more than tripled between September 2024 and September 2025, per Similarweb's Gen AI stats report. A channel growing that fast is worth pulling out of Direct so you can see it. For the fundamentals of how GA4 buckets sessions, see our Google Analytics SEO tracking guide.

What is GA4's native AI Assistant channel, and is it enough?

It is a default channel Google added on May 13, 2026 that auto-classifies recognized AI referrals, and on its own it is not enough.

When an incoming session carries a referrer that matches Google's recognized list, GA4 now tags it with the medium ai-assistant and the campaign (ai-assistant) and files it under the AI Assistant channel with no setup required, as documented by Search Engine Roundtable. You find it under Reports, Acquisition, Traffic acquisition, then set the primary dimension to Session default channel group.

The limits matter. Google's published examples name only ChatGPT, Gemini, and Claude, so Perplexity, one of the highest-intent AI sources, still lands in Referral. The rollout is not retroactive, so it does not backfill your historical data. And like every referrer-based method, it cannot see referrer-less Direct visits. Treat the native channel as a convenient floor, not a complete count, and keep a custom group running alongside it.

How do you build a custom AI channel group in GA4?

Create a channel group, add an AI channel with a Source regex, and move it above Referral.

Here is the sequence:

• Open Admin, then under Data display choose Channel groups, and click Create new channel group.
• Name it something like "SEO plus AI" so it sits next to the default grouping.
• Click Add new channel and name it AI Assistant.
• Set the condition to Source matches regex and paste your AI-domain pattern (below).
• Drag the AI Assistant channel above the Referral channel. GA4 evaluates rules top to bottom, so if Referral fires first your AI sessions stay misclassified.
• Save. The group applies going forward and, unlike a filter, it reprocesses historical sessions in most reports.

A workable Source regex:

chatgpt\.com|chat\.openai\.com|openai\.com|perplexity\.ai|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com|meta\.ai|you\.com|poe\.com|phind\.com

New assistants launch constantly, so treat the pattern as a living list and review it each quarter. If you manage reporting across clients, standardize this group so numbers stay comparable, a discipline we cover in SEO reporting dashboards.

How do you recover AI visits that land in Direct?

You cannot recover the referrer after the fact, but you can build a referrer-based audience and correlate spikes.

Since referrer-less AI visits carry no source, no channel group can retag them. Two practical moves narrow the blind spot. First, build a GA4 audience or exploration filtered on landing pages that only AI answers tend to send traffic to, such as a specific comparison page or a definition you know gets cited, then watch Direct sessions to those pages move in step with your AI Assistant channel. Second, add a UTM habit anywhere you control the link, for example in an llms.txt file or a syndicated answer, so those clicks arrive tagged rather than bare.

The reason this gap is worth the effort: AI referral traffic converts unusually well. Similarweb reported ChatGPT referral traffic converting at 7.1%, second only to paid search at 7.8% and ahead of organic, direct, social, and email, in its Gen AI stats analysis. Undercounting a high-converting channel means undervaluing the content that earns it. To connect this back to money, see SEO ROI measurement.

Native channel vs custom channel group vs referrer audience: which do you use?

Use all three, because each covers a gap the others leave open.

The native channel is your quick floor. The custom group is your working number and your history. The audience layer is where you prove that AI visitors behave differently, which is the argument that gets AI visibility funded internally.

How do you report AI traffic so stakeholders trust the number?

Report it as a labeled channel with a stated method and a caveat about Direct, not as a single headline figure.

In a Traffic acquisition report, switch the channel dimension to your custom group and pin the AI Assistant row alongside Organic Search and Direct. Add a short note in the dashboard that names your method, the date the custom group started, and the fact that referrer-less visits are undercounted. That transparency is what separates a credible AI number from a vanity one, and it protects you when a stakeholder asks why your figure differs from a paid tool's estimate.

Context helps here too. Across independent studies, AI referrals still make up a small single-digit share of total site traffic for most publishers, so a well-built GA4 view usually shows a fast-growing but modest channel rather than a flood. Pair the AI channel with the metrics that already anchor your reporting in our SEO KPI metrics guide, and cross-check against Search Console using the Google Search Console guide for the queries that feed those AI answers.

How does GA4 tracking connect to AI visibility and share of voice?

GA4 measures the clicks AI sends you; visibility tracking measures whether the AI engines mention you at all, and you need both.

Referral analytics only counts people who clicked through. But a growing share of AI answers resolve the user's question inside the response without a click, so a rising presence in AI answers may not show up fully in GA4. That is why AI-era measurement pairs a clicks view (GA4) with a citation and mention view. As a signal of how partial citations still are, Similarweb found ChatGPT included citations in just 0.6% of answers in January 2025, rising to 2.8% by August 2025, a sharp climb that is still a thin slice of all answers, per its Gen AI report.

To measure the half GA4 cannot, track how often the AI engines name you across a fixed panel of buyer prompts, which we cover in AI share of voice, and combine both signals into a single trend line with rank tracking. Disclosure: this guide is written by the team behind Sorank, which builds tooling in this space.

What is the one thing most GA4 tutorials get wrong?

They stop at Source and skip the ordering, the app-traffic caveat, and the historical-data reality.

Most walkthroughs hand you a regex and call it done. Three details decide whether your number is trustworthy. First, channel order: if AI Assistant sits below Referral, GA4's first-match evaluation files your AI sessions as Referral and your custom channel reads near zero. Second, the Direct undercount: because native apps and copy-paste strip referrers, no referrer-based method can be complete, and pretending otherwise invites a credibility hit when your figure looks low. Third, retroactivity: the native channel does not backfill, while a custom channel group reprocesses history in most reports, so the two will disagree for any period before May 13, 2026, and you should expect that rather than debug it.

Name those three caveats in your documentation and your AI traffic report becomes something a finance or leadership audience can actually rely on, which is the whole point of measuring it in the first place.

Frequently asked questions

Related guides