How to Track AI Referral Traffic in GA4 (ChatGPT, Perplexity & More)

Someone asks ChatGPT which WordPress theme to buy, gets a shortlist, clicks your link, and signs up. That visit happened. The open question is whether your analytics can tell you where it came from.

By default, Google Analytics 4 (GA4) files most of these visits under Referral, or worse, under Direct, where they mix with people who typed your URL by hand. Google did add an AI Assistants channel in May 2026, which helps, but it only catches part of the picture.

This guide walks through three setups, from a 30-second check to a permanent custom channel, plus an honest look at what GA4 will never count. By the end, you’ll have a clearer lower-bound estimate of the traffic ChatGPT, Perplexity, Claude, Copilot, and Gemini send your way.

Let’s jump right in! 👇

Why AI referral traffic hides in GA4 🔎

Every visit to your site arrives with two labels: a source (the specific place, like google or chatgpt.com) and a medium (the type of visit, like organic or referral). GA4 uses those labels to sort sessions into channels, the buckets you see in your acquisition reports.

Clicks from AI assistants usually arrive with the medium referral, so for years they vanished into the Referral channel next to random blog mentions. On May 13, 2026, Google added a dedicated AI Assistants channel to the default channel group: when the referrer matches Google’s internal list of AI assistants, GA4 rewrites the medium to ai-assistant and files the session there. The catch? Google doesn’t publish the full list. Its documentation names ChatGPT, Gemini, Copilot, DeepSeek, and Grok as examples, while Perplexity and Claude sessions can still land in plain Referral (check Session source/medium in your own reports to see where yours end up).

Two more leaks matter. Clicks from Google’s own AI Overviews and AI Mode count as Organic Search, not AI Assistants. And when someone taps a link inside a mobile app, or copies a URL and pastes it into a browser, the referrer often disappears entirely. Those sessions show up as (direct) / (none), indistinguishable from people who typed your address from memory.

Here’s the cheat sheet:

AI surface	What usually reaches GA4	Default channel
ChatGPT (web)	chatgpt.com	AI Assistants
Perplexity	perplexity.ai	Referral (usually)
Claude	claude.ai	Referral (usually)
Copilot	copilot.microsoft.com	AI Assistants
Gemini (web)	gemini.google.com	AI Assistants
Google AI Overviews / AI Mode	google	Organic Search
Any AI mobile app	no referrer at all	Direct

One more quirk: the new channel is not retroactive. Sessions recorded before the rollout keep their old labels forever, so your AI Assistants line starts flat in May 2026 even if ChatGPT sent you visitors for years. The fixes below work on your full history. 📊

Step 1: Run a 30-second AI traffic check

Before building anything, see what you already have. Open Reports → Acquisition → Traffic acquisition, switch the primary dimension to Session source, then click Add filter and set Session source → matches regex. Paste this pattern:

chatgpt\.com|chat\.openai\.com|perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com|claude\.ai

That single regex (a regular expression, basically a search pattern) matches the referrer domains of the big five AI assistants plus ChatGPT’s legacy domain. It works on any date range, because GA4 always recorded the referrer; only the channel label is new. Want wider coverage? Extend the pattern with |grok\.com|deepseek\.com|you\.com|poe\.com for the smaller assistants.

One trap to avoid: keep the regex specific. Clever shortcuts like .ai happily match mail, domain, and half the internet. GA4’s custom channel definitions aren’t case-sensitive, but raw UTM parameters are strictly case-sensitive.

If the filtered report shows real sessions, you have proof of life. If it shows nothing, don’t panic: a lot of AI traffic arrives with no referrer at all (more on that below).

Step 2: Build a custom AI Search channel group

The quick filter is great for spot checks, but it disappears when you close the report. For a permanent view that sits next to Organic Search and Referral in every standard report, build a custom channel group. The default group can’t be edited, so you’ll copy it and add one channel of your own:

Go to Admin → Data display → Channel groups.
Open the menu on the Default Channel Group and choose Copy to create new. Name it something like “AI Search view”.
Click Add new channel and name it AI Search.
Add two conditions joined by OR: Session source matches regex (paste the same pattern from Step 1), and Session medium exactly matches ai-assistant. The second condition inherits everything Google’s own list already catches, including assistants it adds later.
Save, then click Reorder and drag AI Search above Referral. Apply and save again.

That last step is the one everyone skips, and it silently breaks the whole setup. GA4 assigns each session to the first matching rule, top to bottom. New channels start near the bottom of the list, below Referral, so a perplexity.ai visit matches Referral first and your shiny new channel stays at zero forever, even though the rule itself is correct.

💡 PRO TIP: Custom channel groups don’t touch your underlying data and they apply to historical sessions, so you can experiment freely. A standard GA4 property allows up to two custom channel groups, with 50 channels in each.

Once saved, your group shows up as a dimension in the Traffic acquisition report. Pick it from the primary dimension dropdown and AI referrals get their own row, trendable over time, comparable against everything else. For the full mechanics, Google’s custom channel groups documentation is worth a skim.

Step 3: Find the pages AI assistants actually cite

Volume is nice. Knowing which content earns the citations is actionable. Head to Explore, start a free-form exploration, and configure it like this:

Rows: Landing page + query string
Values: Sessions (add Engagement rate and Session key event rate if you track conversions)
Filter: Session source matches regex, same pattern as Step 1

The result is a ranked list of the pages ChatGPT, Perplexity, and friends actually send people to. These tend to be comparison posts, buying guides, and pages with very specific details, which is a loud hint about what to write next. 👉 Add Session source/medium as a second dimension if you want to see which assistant favors which page.

The honest part: GA4 undercounts AI traffic

Time for the cold shower. 🚿 Whatever number your AI channel shows, the real figure is higher, for three structural reasons:

No referrer, no record. Mobile apps, in-app browsers, and copy-paste navigation strip the referrer, and those sessions land in (direct) / (none). No regex can match a source field that’s simply empty. ChatGPT softens this by appending utm_source=chatgpt.com to many outbound links, which survives the trip, but it isn’t added to every link.
Google’s own AI answers are invisible. Clicks from AI Overviews and AI Mode arrive with a regular Google referrer and count as Organic Search. There’s no way to split them out inside GA4.
Zero-click answers leave nothing to count. When someone reads your brand name in an AI answer and never clicks, no session exists for any tool to record. That visibility is real, it’s just not in your analytics.

Because those gaps can’t be measured directly in GA4, treat your AI channel as a floor, not a total. The rising trend line matters more than the absolute number.

What to do with the data

Data without a next step is trivia. Three ways to put your new numbers to work:

Compare quality, not just volume. Check engagement rate and key events for AI Search sessions against Organic and Referral. AI-referred visitors arrive pre-sold by the recommendation, so even small numbers can punch above their weight.
Feed the pages that get cited. If a comparison post or guide keeps showing up in your exploration, update it, expand it, and write its neighbors. If some of those pages are aging, this is also a good moment to fight content decay before your citations fade along with your rankings.
Watch the trend monthly. AI referral volume is small for most sites today, but it’s one of the few channels growing this fast. A monthly check-in is plenty; daily refreshing is a hobby, not a strategy. 😅

And if your property was never set up properly in the first place, fix that first: this guide on switching WordPress to Google Analytics 4 covers the whole process, and everything above assumes a working GA4 tag.

Wrapping up 🏁

AI assistants send real visitors, and GA4 finally acknowledges them, but only partially. To see the fuller picture:

🔍 Run the regex filter in Traffic acquisition for instant spot checks.
⚙️ Build a custom AI Search channel group (source regex OR medium ai-assistant, ordered above Referral) for a permanent view.
📄 Use a free-form exploration to find the pages AI assistants cite.
🕳️ Treat every total as a floor: apps, AI Overviews, and zero-click answers never register.

Are you seeing AI referrals in your reports yet, and from which assistant? Let us know in the comments section below!