How can an agency measure GEO and AI search visibility without pretending there is one universal ranking?

Measure a controlled prompt set and the evidence around it instead of presenting one number as a universal rank.

Build a prompt library grouped by intent: category discovery, problem/solution, comparison, local/service-area, brand/entity, and transactional research. Document the exact prompt, platform, date, account state if relevant, and geography or language where applicable.

For each run, capture several dimensions:
- whether the brand/entity is mentioned;
- whether it is recommended, neutrally cited, or merely listed;
- which URLs or sources are cited;
- which competitors appear in the same response;
- whether factual attributes about the business are correct;
- whether the answer changes materially across repeated runs.

Report trends by prompt group, not just an aggregate score. For example: mentioned in 12/20 category prompts, cited directly in 7, correct service-area description in 18, and source coverage concentrated on three URLs.

Pair this with first-party signals: referral traffic from AI products where detectable, branded search movement, assisted conversions, crawl activity, and growth in citations/mentions from sources that AI systems frequently surface.

The goal is repeatable observation, not false precision. A defensible report explains the prompt set, sampling method, date, and limitations so changes over time can be interpreted honestly.