AI referral traffic in GA4: how to see it and why it undercounts

AI referral traffic is the visits that reach your site from ChatGPT, Perplexity, Gemini, Copilot and similar tools. In GA4 you find it by filtering session source for those domains, and you make it permanent with a custom channel group. Whatever number you get is a floor, not the real figure, because a large share of AI-driven visits arrive with no referrer at all. Here is how to set it up in about fifteen minutes, and how to read the result without fooling yourself.

Where AI visits hide in a standard GA4 report

Open Reports > Acquisition > Traffic acquisition and you will not find a row called AI. Visits from AI assistants are filed wherever GA4’s default rules put them, which is usually Referral, and sometimes Direct.

They go to Referral when the browser passes on the page the visitor came from, for example chatgpt.com or perplexity.ai. ChatGPT also adds utm_source=chatgpt.com to many of the links it shows, so those visits carry a source label even when the referrer is missing. They go to Direct when nothing is passed: the link was opened from a desktop or mobile app, copied and pasted, or the visitor’s browser strips referrers.

Change the primary dimension to Session source and type chatgpt into the search box. On most small business sites we look at, there is something there, often more than the owner expected, sitting quietly between Facebook and a directory listing.

Step one: a quick look with a filter

Before building anything permanent, see whether there is enough traffic to care about. In Explore, start a blank free-form exploration, add Session source as a row and Sessions, Engaged sessions and your key events as values. Then add a filter: Session source, matches regex:

chatgpt|openai|perplexity|gemini\.google|copilot|claude\.ai|deepseek|meta\.ai|you\.com

Set the date range to the last 90 days. If you see a handful of sessions, note it and move on. If you see dozens, and especially if any of them triggered a form or a call click, it is worth tracking properly.

One caution about the pattern: keep it to source names you can actually see in your data. Adding broad words like ai on its own will catch unrelated sources and inflate the number.

Step two: make it a permanent channel

GA4 lets you create your own channel group so AI traffic gets its own row in every acquisition report. Go to Admin > Data display > Channel groups, copy the default group, and add a new channel called something like AI assistants. Set its condition to Session source matches regex, using the same pattern as above.

Then the part people miss: drag the new channel above Referral in the list. GA4 assigns each session to the first channel whose rule it matches, so if Referral sits on top, your AI channel will stay empty. Custom channel groups apply to past data as well, so you can compare months straight away.

If you already send lead events to GA4, mark them as key events and you can see AI-referred enquiries side by side with organic and paid. Our AI SEO work in Melbourne starts with exactly this setup, because nobody can tell whether AI visibility is improving without a baseline.

Why the number you see is too low

Treat the AI channel as the minimum. Several kinds of AI-driven visits never show up in it:

• App clicks. Links opened from the ChatGPT desktop or phone app often arrive without a referrer. If the UTM tag was not attached, the visit lands in Direct.
• Google’s own AI. Clicks from AI Overviews and AI Mode are recorded as google / organic, the same as a normal search result. GA4 cannot separate them. Search Console counts them inside its web search data, also without a separate filter.
• No click at all. Someone asks for a good bilingual accountant in Box Hill, reads your name in the answer, then searches your business name on Google a day later. That visit is credited to organic search, or to Direct if they typed your address.
• Consent and blockers. Visitors who decline analytics cookies or run tracking blockers are missing from every channel, AI included.

The third case is the big one for local businesses. The AI tool does the persuading; Google or the address bar gets the credit.

Signals to read alongside GA4

Because the direct number undercounts, look at a few things together:

• Branded search in Search Console. Filter queries containing your business name. Growth in branded impressions, without extra advertising, often reflects people hearing about you somewhere, AI assistants included.
• Direct traffic to deep pages. Direct visits to the home page are normal. Direct visits that land straight on a specific service page or blog article are often pasted or app-opened links.
• Ask people. Add an optional “How did you find us?” field to your enquiry form with ChatGPT or another AI tool as a choice. It is crude, and it is the most honest data you will get.
• Check the answers themselves. Once a month, ask ChatGPT, Perplexity and Gemini the questions your customers would ask, and note whether you are named. We explained what these tools tend to quote in what ChatGPT actually cites.

What to do with the numbers

If AI referrals are close to zero, the job is not tracking, it is being worth quoting: clear service pages that answer specific questions, consistent business details, reviews, and structured data. That is the core of answer engine optimisation. If they are already appearing, look at which pages they land on and build more of whatever is working.

Either way, measure monthly rather than weekly. The numbers are small and noisy, and a single busy week means very little. For a hands-on version of this as part of a broader plan, see our AEO agency service or ChatGPT optimisation, or ask us to check your GA4 setup.

Frequently asked questions