One destination report is a snapshot

What we compared

For each town, we selected its two most recent stored reports as of this review. Each report contains three short lists: places to stay, eat and visit. We trimmed and lowercased names, then divided the shared-name count by the number of distinct names across both lists. A pair with no names would have no overlap score.

This is exact-name comparison. Different spelling can make one place look like two. List truncation, extraction, model behavior and differing underlying answers can all affect the result. The stored aggregates do not retain per-answer success counts or enough model configuration to separate those explanations. The declared model and question counts are metadata, not proof that every call succeeded.

What a destination team can do with this

Keep the dated evidence behind a report and revisit the same visitor questions. A changed list is a reason to inspect the answers and sources. It does not show that a business improved, that a website edit caused the change, or that bookings changed.

Before a member-level study, fix the member sample, prompt wording, model configuration and collection schedule. Retain successful answers and unavailable checks separately. These historical aggregates are useful context; they cannot reconstruct that stronger study after the fact.

Build a scoped destination review or read our measurement methodology.