Two weeks ago I wrote about getting cited by ChatGPT and Perplexity. The obvious follow-up question came back immediately: how do you know if it is working? The honest answer is that the measurement is imperfect, but it is far from impossible, and almost nobody is doing it.
Why the traffic hides
When someone clicks a link inside an AI assistant, the referrer information is often stripped or inconsistent. Your analytics has to put that session somewhere, so it lands in Direct — the same bucket as bookmarks and typed URLs.
This means two things. Your AI-driven traffic is real and it is already arriving. And it is currently invisible, mixed in with everything else you cannot explain.
Setting up the segments
The fix is to isolate the referrers that do come through. In GA4, build a comparison or an exploration filtered on session source containing the assistant domains: chat.openai.com, chatgpt.com, perplexity.ai, copilot.microsoft.com, gemini.google.com.
Save it as a named segment and give it a full year of history. You will almost certainly find it is not zero, and that it has been growing quietly.
Two refinements worth the effort. Watch engagement metrics separately for this segment — AI-referred visitors typically arrive with more context and higher intent than search visitors, and their behavior on site reflects that. And where you control the links, such as your own documentation, add campaign parameters so the attribution survives.
The manual ritual
The analytics side only catches people who clicked. Many buyers read the answer, absorb your name, and arrive later by typing your URL. To see that, you have to check the answers themselves.
Once a month, take the ten questions your buyers actually ask — the ones from your sales calls, not the ones from your keyword tool — and put each to three assistants. Record whether you are mentioned, whether you are cited with a link, and who is named instead of you.
It takes twenty minutes. The competitor column is usually the most useful part: it tells you which sources the models trust in your category, which is a direct roadmap for where to publish next.
What the numbers will look like
Set expectations correctly. As a share of total sessions, this traffic is still small for most sites. Judging it on volume will lead you to dismiss it.
Judge it on trajectory and on quality instead. A segment that is small but doubling, and that converts better than your average, is not a rounding error. It is an early position in a channel that has not yet been bid up.
What to do once you can see it
Look at which of your pages get cited. There is almost always a pattern — usually the ones that answer a specific question early, carry real numbers, and are structured cleanly. That pattern is your template.
Then find the questions where a competitor is named and you are not, and write the page that deserves to be cited instead.
The takeaway
You cannot measure this channel with the precision you are used to from paid search. You can measure it well enough to know whether you are gaining or losing ground, and that is enough to act on. Most of your competitors are not even looking.
