Why AI traffic hides in your reports
Assistant conversations mostly do not pass referrer data: a buyer who gets recommended in a chat and visits later produces a direct session, and one who searches your name afterward lands in branded organic. Some AI surfaces do send occasional referred sessions, Perplexity and ChatGPT link out more than others, but treating those few tagged visits as your AI traffic undercounts it badly; they are the visible edge of a much larger flow. The result is systematic misattribution: channels that deserve credit look flat while direct traffic mysteriously grows, and budget meetings punish exactly the layer that is quietly working.
Building the AI traffic view, practically
Assemble the picture from four signals. First, referrals you can see: create a segment for sessions from assistant domains and AI browsers, small but trend-worthy. Second, branded demand: watch branded search impressions and clicks in Search Console; assistant recommendations show up here first, as people search the name they were just told. Third, direct traffic anomalies: segment new-visitor direct sessions to key money pages, recommended buyers land deep, not on your homepage. Fourth, ask: a one-question how did you hear about us field on your demo and checkout forms, with an AI assistant option, converts anecdote into a measurable series. None of these alone is proof; together, trending monthly, they form a defensible AI traffic line you can put next to any channel, and the upstream side, whether assistants recommend you at all, is what the free and AI share of voice tracking measure.
Connecting the line to revenue
Once the segment exists, treat it like a channel. Tag the self-reported AI arrivals in your CRM and follow their close rate and deal size; teams doing this consistently find assistant-referred buyers convert unusually well, they arrive pre-sold by a recommendation they trust. Reconcile monthly: your share of assistant recommendations, from your visibility tracking, should lead your AI traffic line by weeks, and the traffic line should lead its revenue by your normal cycle. When the chain moves together, you have what most competitors lack entirely: an attribution story for the channel everyone else calls unmeasurable, and a defensible answer to what did the AI visibility budget buy, which pairs with the method in how to know if your GEO is working.
Setting up the segments this afternoon
The whole view takes about an hour in a standard GA4 and Search Console setup. Create one exploration segment for sessions whose source matches the assistant domains you care about, and annotate today's date so the trend has a birthday. Add a comparison segment for new-user direct sessions landing on your five money pages, that is your recommended-buyer proxy, and check it monthly, not daily. In Search Console, bookmark the branded-query filter and record this month's impressions as the baseline. Then add the how-did-you-hear field to your two highest-intent forms with a stable option list, and tag arrivals in the CRM from day one. None of this requires new tooling, and a month from now you will have the first honest reading most competitors never build: how much of your pipeline the assistants are quietly feeding.
