A known referral is observable traffic. AI influence is a broader, less directly observable question.
Define a referral consistently
An AI referral is a visit your analytics identifies as arriving from an assistant or related surface. Start with the source information your analytics actually receives. Preserve unknown and direct traffic as their own categories; renaming all unexplained visits ‘AI’ produces an attractive report without evidence.
Maintain a reviewed list of identifiable sources and document the matching rule. Include the full source or referrer where your analytics and privacy policy permit. Do not classify by a fragment so broad that unrelated domains become counted as assistant traffic.
Validate the measurement journey
Test an ordinary public landing-page visit from the relevant surface where possible. Check that the page loads, the analytics event fires according to consent settings, and the recorded source is what you expect. A broken event or an intervening redirect can look like a channel-performance problem.
Inspect landing pages as well as totals. Product, documentation, pricing, and editorial pages attract different tasks. Grouping them together can conceal whether people are arriving to buy, troubleshoot, or simply read a definition.
Connect traffic to useful outcomes
Report observed sessions, engaged visits under your analytics definition, completed goals, and qualified downstream outcomes. Keep the conversion window and identity rules consistent. Exclude internal testing and obvious automated requests where your system supports a reliable method.
Use sample size to guide interpretation. Two purchases from ten known referrals are interesting, but do not establish a durable channel conversion rate. Show the counts and date range alongside any percentage.
Keep influence as a separate question
Someone may read an AI answer, remember a brand, and visit later through another route. Referral analytics alone cannot capture that journey. Customer interviews and a neutral ‘how did you hear about us?’ question can add context, although self-reports have their own limits.
Use three labels in reporting: observed referral, self-reported influence, and unconfirmed hypothesis. This keeps the team open to a changing discovery journey without converting uncertainty into attributed revenue.
