Do your prospective buyers actually understand what you do? Find out free →
Attribution guide

How to Measure AI Referral Traffic and Customers

A practical measurement model for separating AI crawler requests, answer retrievals, referral sessions, conversions, and assisted revenue.

Direct answer

Measure AI as a funnel, not one traffic number. Keep crawler requests separate from human referral sessions; preserve referrer, landing page, and campaign parameters; mark meaningful actions as conversions; and report first-touch and assisted outcomes alongside answer visibility. A crawler request is evidence of access, not a visit or customer.

Evidence from Optimly's public index

Why a zero referral count can coexist with heavy crawler traffic

Optimly's 2026-W32 report recorded 146,188 agent visits but zero referrals specifically attributed to AI Answers. That is methodologically possible: crawler and retrieval logs observe machine requests, while referral analytics require a person to click through with identifiable source information. Some AI products omit referrers, so part of the traffic can also appear as direct.

Inspect the source report

Use five separate stages

Do not combine machine access and human acquisition. Give every stage its own event and denominator.

  • Crawler access: verified AI crawler requests by bot and URL.
  • Answer retrieval: user-directed agents fetching a page during an answer workflow.
  • Referral session: a human landing from an identifiable AI product or tagged link.
  • Conversion: a business check, signup, booked meeting, or purchase.
  • Assisted outcome: a later conversion from someone whose earlier session came from AI.

Preserve source information

Capture the document referrer before navigation changes, classify known AI domains, retain UTM parameters on links you control, and persist first-touch AI attribution across the session. Keep an explicit unknown or direct bucket rather than forcing unattributed traffic into AI.

Report the landing page that earned the visit

Group referral sessions and conversions by the first content path: comparison, methodology, brand profile, guide, or case study. This reveals which content creates commercial movement instead of merely attracting crawlers.

A practical starting plan

  1. 1Create a maintained list of AI referral domains and campaign values.
  2. 2Store first-touch source, landing path, and timestamp before the user moves into the app.
  3. 3Define one activation event and one qualified conversion event.
  4. 4Build a funnel from referral session to activation to qualified outcome.
  5. 5Reconcile analytics with server logs weekly, but never treat crawler counts as human sessions.

Frequently asked questions

Does ChatGPT always pass a referrer?

No. Referral information can be missing or altered, so analytics should retain an unattributed bucket and use tagged links where you control the destination.

Are ChatGPT-User requests referral sessions?

No. They indicate a user-directed agent retrieved a page; they do not prove the person clicked through to the website.

What should count as an AI conversion?

Use the same meaningful outcome used for other channels, such as a completed business check, signup, qualified meeting, or purchase, and attach the AI source rather than inventing a weaker event.

Sources and methodology

Optimly publishes this guidance and has a commercial interest in AI brand measurement. The dated evidence and external methodology sources below make the claims independently inspectable.

See how AI understands your business

Create your Business Profile, check what AI gets right or wrong, and establish a baseline before you change anything.

Check my business — free

Continue evaluating