AI-search attribution connects a captured enquiry to the evidence available across an AI answer, a referral or visit, a landing page, and a form submission. It is harder than ordinary campaign attribution because AI assistants do not expose a uniform click or prompt identifier.

A defensible model therefore reports confidence levels and preserves the evidence behind them.

Four signals that should remain separate

Answer evidence

Did the brand or managed page appear in a monitored answer before the enquiry? Preserve the prompt, engine, timestamp, answer text, position, and sources.

Referral evidence

Did the browser provide an identifiable AI referral? Some visits carry a useful referrer; others arrive through copied URLs, privacy-protected apps, or direct navigation. Missing referral data does not prove that AI had no influence.

Landing-page evidence

Which page did the visitor enter, and was it a page created for the monitored buyer question? This connects the visit to a publishing hypothesis even when the exact prompt is unavailable.

Enquiry evidence

Did the same session submit a form, and did the submission meet the team's qualification rules? A page view is not pipeline. The enquiry record needs its own timestamp, form, page, and consent context.

A practical confidence scale

Use a high-confidence label when the answer capture, identifiable AI referral, managed landing page, and enquiry event align. Use medium confidence when the answer and landing page align but the referrer is missing. Use influenced or unverified when the evidence is suggestive but incomplete.

The purpose is not to force every enquiry into a perfect source bucket. It is to show the strongest supported relationship without hiding uncertainty.

What the report should contain

For each attributable or potentially influenced enquiry, store:

enquiry and session identifiers;
landing page and first-seen time;
available referrer and campaign parameters;
relevant monitored prompt and engine captures;
qualification status;
attribution confidence;
limitations or conflicting signals.

Aggregated reporting can then show enquiry volume by page, engine evidence, confidence level, market, and sales outcome.

Avoid these attribution mistakes

Do not label every direct visit to a managed page as an AI lead. Do not use an AI crawler visit as proof that a buyer was referred. Do not claim an engine caused an enquiry merely because the brand appeared in a recent answer. And do not combine visibility lift with revenue as though they are the same metric.

Connect attribution to publishing decisions

The point of attribution is operational. If a page gains citations but produces no qualified visits, its conversion path may need work. If a page produces enquiries without monitored visibility, another acquisition path may be responsible. If neither signal moves, the underlying buyer question or evidence strategy should be reconsidered.

This measurement discipline is the DELIVER stage of AI-search lead generation for B2B. See how TopGeo reports results for the full evidence contract.

AI-Search Attribution for B2B: A Defensible Measurement Model