How do you track traffic and referrals from AI answer engines?

GA4's default channel group includes a channel called AI Assistant, for users who arrive "from sources like ChatGPT, Gemini, Deepseek, Copilot, or Grok." Analytics classifies those sessions itself, so a property that's already collecting data has the channel without a custom rule.

Read it at session scope. The session default channel group describes how each new session began, and the first user version describes how a person first reached you. A question about last month's visits is a session question.

The channel excludes Google's AI Overviews and AI Mode, which Google files under Organic Search. So the number counts visits from assistant products and says nothing about Google's AI results.

To see which assistants sent the visits, add session source and medium to the report. Source names the referring site or platform. Read the rows before you total them, in case one assistant appears under more than one source name.

Default channels can't be edited. A custom channel group is optional, for the case where an assistant you care about is being classified somewhere else.

HubSpot's Sources report has its own category, AI Referrals, available with Marketing Hub or Content Hub at the Professional and Enterprise levels. HubSpot checks its source rules in order. A session reaches AI Referrals if an earlier rule, such as paid search or email marketing, hasn't already claimed it, and if the referring domain or a UTM parameter points to an AI platform HubSpot recognizes. The list covers ChatGPT, Claude, Perplexity, Microsoft Copilot, Google Gemini, Meta AI, Mistral, Poe and Grok. Click the Sessions number beside AI Referrals to see the total by platform.

Like Analytics, HubSpot keeps regular Google and Bing searches in Organic search. Only an AI-specific domain such as gemini.google.com counts as an AI referral.

The two tools keep separate lists and apply their own rules, so their totals for the same month can differ. We'd pick one as the reported figure and keep the other as a cross-check. If your contacts and deals are in HubSpot, its number is the easier one to carry into the CRM step.

Google's AI Overviews and AI Mode are reported in Search Console, in the generative AI performance report. Google made it available to all websites as of August 31, 2026. It counts impressions, meaning the number of times links to your site were shown to a user in one of those features. You can group them by page, country, date or device.

An impression is a display. It isn't a visit, and the report offers no click count or query list for these features. The clicks do reach your site. Analytics counts them in Organic Search with clicks from ordinary results. So Search Console tells you which pages Google's AI features are showing. It can't tell you how many visits those features sent.

The report covers Google Search only. The Gemini app is a separate product, and a visit from it belongs with the assistant sessions above. If the report is missing from your property, Google says the usual cause is too few impressions in these features. It can also mean the site has been excluded from them in Search Console.

It's tempting to treat a rise in direct traffic as assistant visits that lost their label. Analytics doesn't support that reading. Google describes direct traffic as visits with no clear referral source, and it lists ordinary causes: links without tracking parameters, redirects, URL shorteners, ad blockers, typed addresses, and links inside PDFs or Word documents.

An assistant visit that arrived with no usable source information would land in that same group, and nothing in the report would set it apart from the rest. That means the size of any unlabeled share is unknown, and no formula turns direct traffic into an AI estimate. Report the assistant sessions you can see as observed visits, and leave the direct line as it is.

A session count doesn't tell you whether assistants are sending buyers. The CRM gets closer, because a contact record carries a source and a deal has an outcome.

Lean Labs' AEO service shows one way to lay the report out. It describes three groups of measures. Visibility holds brand mentions, citations and prompt coverage. AI referral traffic and engaged visits sit under demand, and contacts, deals and pipeline sit under revenue.

The visibility group is where prompt and citation checks belong. A check records whether an assistant named or linked you when it was asked a set question. That's a sample of answers and it counts no visitors, so it sits with the Search Console impressions and apart from traffic. Repeat the same prompts on a schedule, because one reply is a single sample. You can run the checks by hand or pay for a tool that runs them, and we price those tools in how much an AEO audit costs for a B2B website.

Suppose a quarterly report shows 12,000 impressions in Search Console's AI report and 400 assistant sessions in analytics. Suppose the CRM shows nine new contacts with an AI referral as their recorded source, and two deals associated with those contacts. Each figure is true for its own system, and they don't form a funnel.

The impressions come from Google's AI results and the sessions come from other assistants, so dividing one by the other gives a rate that describes nothing. Sessions are counted per visit. Contacts are counted per person, by the source the CRM recorded. Those two totals don't reconcile automatically. The two deals are linked to contacts that have an AI source. That link shows the source the CRM recorded for the contact. It doesn't show that the assistant caused the sale.

The report is still useful once each number keeps its label. You can say which pages Google is showing in its AI results, and how many visits assistants sent. You can say how many contacts carry an AI source and how many deals are attached to them. The CRM counts can trail the others, because a contact or a deal can be recorded later than the visit that began the sales conversation, a lag we work through in how long before AEO generates qualified leads.