How to Track AI Referral Traffic in GA4: ChatGPT, Perplexity, Gemini, and Claude

To track AI referral traffic in GA4, you need to identify sessions where the referrer domain matches a known AI assistant, such as chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, or copilot.microsoft.com, and group them into their own channel instead of letting GA4's default rules file them under Referral, Direct, or Organic Search. GA4 added a native AI Assistant channel in May 2026 that does some of this automatically, but it has two significant gaps: it excludes Perplexity entirely, and it cannot see any session that arrives without a referrer at all, which is common when a link is opened from inside an AI app's in-app browser.

For a site built around getting cited by AI answer engines, this matters more than it would for an ordinary marketing site. If your GA4 reports quietly fold ChatGPT and Perplexity sessions into Direct traffic, you cannot demonstrate that a citation drove a real visit, cannot compare AI-referred engagement against organic search, and cannot show which landing pages are actually converting AI mentions into pipeline. Fixing the attribution takes about fifteen minutes.

Why GA4's default channel grouping misses most AI traffic

GA4's default channel grouping was built before AI answer engines existed as a meaningful traffic source, and it still reflects that. A session with a referrer of chatgpt.com and no UTM parameters is classified as Referral, sitting in the same bucket as a random blog linking to you. A session that arrives with no referrer header at all, which happens whenever a user copies a link out of a chat response or opens one from a native mobile app, is classified as Direct, indistinguishable from someone typing your URL from memory. Both outcomes erase the fact that an AI answer is what actually sent the visitor.

The quick method: GA4's native AI Assistant channel

As of the May 2026 update, GA4 ships a built-in AI Assistant channel that automatically catches sessions referred from ChatGPT, Gemini, and Claude when a referrer is present. To check whether it is already picking up your traffic, open Reports > Acquisition > Traffic acquisition and look for AI Assistant in the Session default channel group list. If you see it with sessions attached, the native rule is working for at least part of your AI traffic.

The native channel has two blind spots worth knowing before you rely on it. First, it does not include Perplexity, which is one of the highest-intent AI referral sources for most B2B and technical content sites. Second, like every channel rule, it depends on a referrer header being present, so it inherits the same undercount problem as the rest of GA4 for in-app and copy-paste traffic. Both gaps are reasons to build a custom channel group rather than stop at the default.

Building a custom channel group that catches everything

A custom channel group lets you define your own matching rule and place it ahead of GA4's built-in rules, so traffic that would otherwise fall into Referral or the incomplete AI Assistant bucket gets classified correctly first.

1. In GA4, go to Admin > Data display > Channel groups
2. Select Create new channel group and name it something identifiable, such as "AI Referral Sources"
3. Add a new channel rule and set the condition to Session source contains, then enter a regex matching the assistants you care about: chatgpt\.com|chat\.openai\.com|perplexity\.ai|claude\.ai|gemini\.google\.com|bard\.google\.com|copilot\.microsoft\.com|meta\.ai
4. Name the matched channel "AI Assistants" so it reads clearly in reports
5. Drag your new AI Assistants rule to the top of the rule list, above the default Referral rule
6. Save the channel group, then apply it as a comparison or secondary dimension on your Traffic acquisition report

Backfilling and inspecting historical AI traffic

A new channel group only reclassifies data going forward from a small lookback window, so it will not retroactively relabel months of past sessions. To see how much AI traffic you were already getting before you built the channel group, use an Explore report instead of the standard channel view.

• Go to Explore > Free-form and start a blank exploration
• Add Session source / medium as a dimension and Sessions, Engaged sessions, and Average engagement time as metrics
• Apply a dimension filter using the regex from the channel rule above to isolate matching rows
• Sort by Sessions descending to see which assistant is sending the most traffic historically
• Add Landing page as a secondary dimension to see which pages each assistant is actually sending visitors to

This Explore view is also the fastest way to sanity-check a new channel group before you trust it in a dashboard. If the regex is wrong, sessions you expect to see from perplexity.ai or chatgpt.com simply will not appear in the filtered table, which is much easier to spot in a free-form report than in an aggregated channel chart.

What to do once AI referral traffic is visible

Isolating the channel is the setup step, not the goal. Once AI referral sessions are broken out cleanly, three comparisons become possible that were not before. Compare engaged session rate for AI Assistants against Organic Search on the same landing pages, since AI-referred visitors typically arrive with a specific question already answered and often show different engagement patterns than someone still evaluating options from a search results page. Compare which landing pages receive AI referral sessions against which pages Google Search Console shows appearing in AI Overviews, since a page can show impressions in one surface and clicks in the other without the two lining up. And track the trend over time per assistant, since a rising ChatGPT channel and a flat Perplexity channel calls for a different content response than the reverse.

Feed what you learn back into your measurement stack rather than treating it as a one-off report. If you are already running the broader AEO metrics programme this site recommends, the GA4 AI Assistants channel becomes one more input alongside citation frequency and share of voice, and it is the one input that ties an AI citation directly to a session, a landing page, and downstream conversion behaviour.

Common pitfalls that undercount AI traffic further

Even a correctly built channel group leaves real gaps, and knowing them prevents you from over-trusting the final number.

• Dark traffic from copy-paste behaviour: a user who reads an answer in ChatGPT and then manually types your URL, or copies a link into a new tab, generates a Direct session GA4 cannot connect to the AI answer that caused it
• In-app browser stripping: opening a link from inside the ChatGPT or Perplexity mobile app can drop the referrer entirely, depending on the app version and platform, again landing the session in Direct
• Self-referral pollution: if your own domain is not in GA4's referral exclusion list, cross-subdomain navigation can occasionally be misattributed as a Referral session, diluting the signal in the channel you are trying to isolate
• New assistants and rebrands: the regex above covers the assistants with meaningful traffic today, but AI search products launch and rename often enough that the rule needs a periodic review, not a one-time setup

GA4 alongside Search Console, not instead of it

GA4's AI Assistants channel tells you who actually clicked through to your site. It has no visibility into citations where an AI engine answered a question using your content without sending a click at all, which is the majority outcome for most AEO efforts. Search Console's AI Overviews reporting and a direct AI citation gap analysis against competitors cover that half of the picture. Use GA4 to measure what happens after someone clicks, and the other two to measure whether you are being cited in the first place.