Measuring AI Referral Traffic in GA4: A Practical Guide
How to separate visitors from ChatGPT, Perplexity and similar tools in GA4: referral sources, a custom channel group and the limits of the data.
rabbitclip teamPublished: 6 min read
Short answer
AI referral traffic is made up of visitors who reach your site by clicking a link inside an AI assistant's answer, and GA4 does not label it as a separate channel on its own. These visits are usually recorded as referrals, so the session source is the domain of the assistant. You measure them by filtering those sources and grouping them in a custom channel group that you can check regularly. The measurement is always incomplete, though: answers nobody clicks and visits that arrive without a source never reach any report.
This article is a companion to the one on measuring visibility in AI search. That piece asks whether you are mentioned in answers; this one asks how many visits those mentions send to your site. The two are not the same thing and should be read separately.
Where does AI traffic show up in GA4?
GA4 sorts traffic by the source and medium of the session. When a visitor clicks a link they saw in an AI tool, the browser often passes that tool's address as the referrer. GA4 records it as a referral and the tool's domain appears in the source column. To see these names on your own site, open the traffic acquisition report and look at the source dimension.
Which domains appear depends on the tool and the date. You may see sources resembling ChatGPT, Perplexity, Gemini or Copilot, but the list is not fixed: tools change addresses and new ones arrive. It is safer to open the source list in your own data and look for names you recognise than to copy a ready-made list from the internet.
Reading single source rows is untidy work. One tool can have several domains or subdomains, and you have to add them up by hand to see a total. That untidiness is why most teams never separate AI traffic at all, and it stays buried inside general referrals.
How do you set up a custom channel group?
GA4's default channel groups have no separate channel for AI. The fix is to define your own custom channel group and collect the sessions whose source matches certain patterns under a channel called 'AI assistants'. The channel is defined on the source dimension, using sources that contain particular terms or match a regular expression. The admin screens change often, so I will not describe the buttons one by one; GA4's own help pages give the current steps.
A few points to watch while setting it up:
A channel group may not apply retroactively to old data; it works reliably from the day you create it. The earlier you set it up, the longer the comparison period you build.
- Build the definition from sources you actually see in your report, not from guesses
- Review the rule every three months, as new tools appear
- Place the rule above the general referral rule so that one does not claim the sessions first
- Give the channel one plain name that everyone on the team will understand
Which metrics should you read this traffic with?
Once the channel exists, the first thing to look at is quality, not volume. AI referral traffic is small on most sites, and that is normal. The real question is what that small group does on your site: how many pages they view, which page they land on, and whether they fill in a form or complete another goal such as a purchase.
Put organic search visitors next to them for comparison. Seeing the conversion of both channels on the same goal tells you whether AI traffic is useful to you. Working out percentages on a tiny sample is misleading, so say plainly in the report that the numbers are small.
What mistakes are common in this measurement?
The most common mistake is treating the number in GA4 as the whole of your AI visibility. The number shows only people who clicked and whose source was passed on. If your brand was named in an answer and nobody clicked, or a reader remembered the name and typed the address later, that effect never shows up as an AI channel in GA4.
The second mistake is reading meaning into a day or a week of movement. In a low-volume channel, three visits in two days can pull the chart upwards. Read the trend over a window of at least a few months.
The third mistake is setting the measurement up once and leaving it. If you do not check regularly that the tag still fires and the channel rule still catches the right sources, your report quietly goes incomplete.
- Presenting AI traffic as the whole of visibility
- Writing commentary on short-term fluctuation
- Writing the channel rule once and never reviewing it
- Assuming one tool's address is all AI traffic
Which visits can never be measured?
Some visits arrive without a source by their very nature. If an AI app opens a link inside itself or in a way that does not pass the referrer, GA4 counts the session as direct. Clicks from the AI summaries in Google's search results are generally not distinguishable from ordinary Google search either. There is no reliable way to know how much of that traffic came from AI.
This limit is not a defect; it is the nature of the measurement. You can put UTM tags on your own links, but you do not decide which address an AI tool gives in its answers. So the correct sentence for a report is 'the AI traffic we can measure', not 'total AI traffic'.
When is this measurement not worth the effort?
If your site has very few visitors overall, building and reporting a separate channel is not worth the effort; the numbers are too small to read anything from. First make sure basic conversion tracking works, then move on to this split. In an account with no goals defined, an AI channel tells you nothing either.
The measurement is also pointless if it will not change what you do. If you will not alter content, brand consistency or page structure after seeing the number, the report is only decoration.
Measuring AI referral traffic comes down to seeing the sources in GA4, grouping them in one custom channel and reading the result next to organic traffic, always knowing the picture is incomplete. Even when the number is small, its quality, the pages it lands on and what it completes are useful for decisions. If you would like to think through how to set this up on your own site, we can arrange a discovery call with rabbitclip.
FAQ
Can I separate visitors from ChatGPT in GA4?
Largely yes. Visits that pass their source are recorded as referrals and appear in the source dimension, and you can group them in a custom channel. Visits that do not pass a source mix into direct traffic and cannot be separated.
Does GA4 separate AI traffic automatically?
No. The default channel groups contain no separate channel for AI. You have to define it yourself, and the definition needs updating as new tools appear.
Do visits from Google's AI summaries show up separately?
Generally they do not. Those clicks are mostly counted within normal Google organic search. GA4 alone cannot fully measure the effect of AI summaries.
My AI traffic is very low. Should I worry?
On most sites this channel is small, and that alone is not a problem. What matters is less the volume than what these visitors do on the site and how it changes over time.
Do I need an extra tool for this measurement?
No, GA4's own custom channel group feature is enough. If you want deeper source analysis you can also look at server logs, but it is not needed to begin with.
