Public methodology
How AgentLocal measures local AI visibility
AgentLocal combines attributed public business facts with repeatable test questions. We report what assistants returned, which public links appeared, and whether the requested constraint was answered. We do not claim access to private user prompts or hidden ranking systems.
Reviewed
What we collect
- Business facts
- Names, locations, services, hours, contact methods and other details supported by a public source.
- Test questions
- Controlled, non-private queries representing real local-service intents and constraints.
- Assistant outputs
- Whether an answer addressed the constraint, named an identifiable business, and included a source link.
- Freshness
- When a fact was observed and when a resolution page last passed its evidence-quality gate.
How to read the numbers
Every rate includes its numerator and denominator. A citation is a public URL present in a measured answer; it is not proof that a model relied on that page internally. Results are observations from a defined window and test corpus, not a promise of future ranking.
- Unmeasured markets are labelled unmeasured, never scored as zero.
- Repeated links in one answer count once for answer-level citation rate.
- Commercial plans do not change organic ordering or measurement.
- Private prompts and raw IP addresses are not published.
Publication gates
A local guide is published only when it has observed demand, multiple eligible businesses, current evidence, enough constraint coverage and at least one current handoff. Pages that fall below the gate are removed from public indexing until they recover.