Spoke
Segmenting AI referrers
The visitors who do click from an AI answer are identifiable in ordinary analytics. It is the easiest measurement win available here, and it counts the smallest part of the channel.
That ChatGPT tags its referrals with an identifiable source parameter is documented by OpenAI. The segmentation technique is standard analytics practice.
Some people do click. Those visits arrive with a referrer or a tagged parameter, which makes them segmentable with no new tooling — and getting this set up takes about an hour.
How do you track traffic from ChatGPT and other AI assistants?
Create a segment or filter matching the assistant domains in your referral data. OpenAI documents that ChatGPT referrals carry an identifiable source parameter, which makes those visits trivially separable; other assistants generally arrive as ordinary referrals from their own domains.
What this measures is people who clicked through. It does not measure the larger group who read an answer citing you and never visited, which is the gap the attribution spoke covers.
Setting it up
Identify the sources. The assistant domains you care about, plus the tagged parameter OpenAI documents. Keep the list in one place; it grows as products launch.
Create a segment, not just a filter. You want to compare these visits against organic and direct on the same metrics, which needs them as a comparable group rather than an isolated view.
Separate it from organic search. Depending on your analytics, assistant referrals may land in referral, in direct, or in a bucket of their own. Leaving them mixed into organic makes both numbers wrong.
Check it monthly. New products appear, referrer behavior changes, and a segment built once and forgotten quietly stops matching.
What to look at once it exists
Resist reporting the visit count as the headline. It is the smallest and least interesting thing in the data.
Behavior on arrival. These visitors frequently arrive further along than organic ones — they have already had the question answered and are checking you specifically. Pages per session, conversion rate and the pages they land on all tend to differ, and that difference is the story.
Conversion rate against other channels. If it is meaningfully higher, that is an argument for the channel that does not depend on volume — which matters, because the volume will look small.
Which pages get the click. The pages that earn a visit after the answer was already given are your most persuasive pages. That is useful information for content planning and it is not available anywhere else.
The limitation, stated plainly
This counts clickers. If the channel works the way this pillar argues, most of its value is in people who never clicked.
So a segment showing small numbers is not evidence the channel is small. It is evidence that this instrument measures the part that produces sessions, which is the part that was never the problem.
Use it for what it is good at — showing that the visits you do get from assistants behave differently and usually better — and use the prompt panel and self-reported attribution for the rest.
More in Measurement
Prompt panel tracking
Asking an assistant about yourself once tells you nothing. A fixed set of prompts, re-run on a schedule and recorded, tells you a direction — which is the most that is available.
AI visibility reporting
Google and Microsoft now report AI visibility directly, and OpenAI tags its referrals. That beats a third-party citation tracker — and each one measures something narrower than its name suggests.
The attribution gap
Being cited without being clicked is a real outcome that last-click reporting cannot see. It usually resurfaces later as direct or branded traffic, credited to the wrong thing.
AI share of voice
How often you are named across a fixed set of questions, relative to the people you compete with. It is noisy, it is the closest thing available, and it breaks the moment anyone reports it as a single number.
Self-reported attribution
One required field on every intake. It is biased, imprecise, and the only place a human tells you directly that a machine recommended you.
KPIs that survive
Sessions as a headline metric tells a story that gets less true every quarter. These are the measures that keep meaning what they meant when the traffic mix changes underneath them.