Tracking AI Search Referrals in GA4: See Traffic From ChatGPT

Where AI assistant traffic hides in GA4

When someone asks ChatGPT for a recommendation and clicks through to your website, GA4 records the visit, but it does not label it helpfully. By default, clicks from ChatGPT, Perplexity, Claude and Microsoft Copilot land in the Referral channel, mixed in with links from directories, partner sites and forums. A portion arrives as Direct or Unassigned when the referrer gets stripped. Nothing appears under a tidy AI label, because Google's default channel grouping simply has no such channel.

The practical consequence is that plenty of UK small businesses already receive traffic from AI assistants and have no idea. The fix is a custom channel group: you tell GA4 which referral sources belong to AI tools, and that traffic gets its own line in your acquisition reports. The whole job takes well under an hour and needs no code changes on your website.

The referral sources worth tracking

Each assistant passes an identifiable source when a user clicks a link inside its answer. The list evolves as products change, but these are the mainstays:

• chatgpt.com — ChatGPT's domain. It also appends utm_source=chatgpt.com to many outbound links, which makes it the easiest assistant to identify. Older data may still show chat.openai.com
• perplexity.ai — Perplexity cites sources prominently in every answer, so it tends to send more clicks relative to its user base than other assistants
• copilot.microsoft.com — Microsoft Copilot; you may also see edgeservices.bing.com from older Copilot surfaces
• gemini.google.com — the Gemini app and website
• claude.ai — Claude

Before building anything, check what you already receive. Open Reports, then Acquisition, then Traffic acquisition, switch the primary dimension to Session source / medium and search for "gpt", "perplexity" and "copilot" in turn. Note the exact source strings you find; they belong in your regex.

One caveat to accept up front: clicks from Google's AI Overviews and AI Mode arrive labelled as google / organic. GA4 cannot separate them from classic organic search, so this tutorial covers the standalone assistants only.

Build a custom channel group step by step

• 1. In GA4, open Admin, and under Data display choose Channel groups
• 2. Click Create new channel group. GA4 copies the default grouping as your starting point; name it something like "Channels incl. AI"
• 3. Click Add new channel and name it "AI Assistants"
• 4. Set the condition to Source matches regex, then paste the pattern from the next section
• 5. Drag the AI Assistants channel above Referral in the list. Order matters: GA4 assigns each session to the first channel it matches, so if Referral sits higher it will swallow your AI traffic
• 6. Save, then open Traffic acquisition and switch the primary dimension to your new channel group

Custom channel groups are evaluated when reports run, so your AI Assistants channel applies to historical data as well. You can see the trend from when the property started collecting, not just from today. Your default channel group stays untouched, which means nothing breaks for colleagues who rely on the standard view.

A regex that catches the traffic

This pattern covers the major assistants as of mid-2026: .*(chatgpt\.com|openai\.com|perplexity\.ai|copilot\.microsoft\.com|edgeservices\.bing\.com|gemini\.google\.com|claude\.ai).*

Two details matter. First, GA4's channel conditions require a full match, which is why the pattern starts and ends with .* rather than matching a bare domain. Second, the source dimension respects UTM parameters, so a link tagged utm_source=chatgpt.com is caught by the same pattern even when the referrer was stripped. Revisit the regex quarterly: new assistants appear, domains change, and five minutes of maintenance keeps the channel honest.

Go deeper with Explorations

The channel group covers standard reports. For real insight, build a free-form exploration: open Explore, choose Free form, add a filter where Session source matches your regex, set Landing page + query string as the row dimension, and use Sessions and Key events as values. This answers the most useful question in AI search: which of your pages do assistants actually cite?

For most sites the answer is rarely the homepage. Assistants link to pages that answer questions directly: guides, FAQs, pricing explanations and comparison pages. If your exploration shows a handful of guide pages earning nearly all AI referrals, that is your evidence for writing more of them. Add a second breakdown by session source to see whether ChatGPT and Perplexity favour different pages, because they often do.

What counts as a normal share?

Honest benchmarks first: for most UK small-business websites, AI assistant referrals are still a small slice of total traffic, typically low single digits as a percentage of sessions. Sites with strong informational content, B2B firms, and anyone frequently cited in comparison-style answers see more. The share has climbed steadily since 2024, and the direction of travel matters more than this month's number.

Judge the channel on quality, not volume. An AI referral often arrives pre-qualified: the visitor has already read a summary of what you do and clicked through deliberately. Compare engagement rate and key-event conversion against your organic channel before deciding whether the traffic deserves more effort. Small volumes that convert well justify attention; large volumes that bounce do not.

Blind spots to keep in mind

This setup measures clicks, and clicks are only part of the picture. It cannot see answers where users never click at all. It cannot separate AI Overview clicks from ordinary Google organic. Mobile apps sometimes strip referrers, so a share of assistant traffic hides inside Direct. And crawler visits from AI bots fetching your content never register as sessions; server logs show those, GA4 does not.

Treat GA4 as one instrument, paired with a monthly audit of what assistants actually say about your brand, and together they give a workable view of AI search performance. If you would like the tracking configured and benchmarked properly, our team can set it up alongside your wider analytics.