How to Measure a Sample of AI Search Visibility
Measurement should describe a specified response sample, not all assistant users or a platform’s market. Begin with a written protocol and preserve the underlying observations.
Define the sample
Document why each question was selected and classify its audience, topic, and intent. Freeze the set for comparisons. State platform, interface, displayed mode, region, language, account state, and collection window.
Use observable fields
- Brand mentioned: yes or no under a written matching rule.
- Displayed source link: URL and domain as shown.
- Factual accuracy: coded against a named public source.
- Position or sentiment: only with a reproducible coding rule.
- Referral session or conversion: only under the documented analytics model.
Calculate transparently
For every rate, publish the numerator and eligible denominator. Explain exclusions, failed collections, duplicate prompts, and repeated runs. Keep platform results separate when interfaces or collection conditions differ. Do not hide unfavorable observations inside a composite score.
Compare with discipline
Use the same questions and collection process for a repeat wave. Log site releases, paid campaigns, publicity, seasonality, and platform changes that could affect interpretation. A before-and-after difference alone does not establish causation.
Limits and next step
Sampling error, output variation, personalization, and interface changes limit conclusions. Report the result as a dated sample and retain the captures for review.