Results are controlled samples

Each result comes from a defined prompt, platform, location, language and session type, run on a schedule. It describes what an assistant recommended at that moment under those conditions — not a permanent or universal ranking.

AI answers vary

Different runs, interfaces, models, locations and phrasings can produce different recommendations. That variation is expected and is why we repeat priority prompts and report distributions rather than single answers.

Ranking represents recommendation order

We extract every business recommended in an answer, resolve aliases and spelling variants, remove duplicates, and record the order in which unique businesses appear. Your position is your first appearance in that de-duplicated order.

Extended rankings

If a business does not appear in the initial answer, the same conversation is continued with a neutral request for ten further distinct recommendations, repeated in batches until the business appears, the configured depth is reached, or the model stops providing valid new names. Extended positions are labelled as such.

Screenshots represent a specific dated run

Every screenshot is tied internally to the exact prompt, conversation, timestamp, platform and collector settings. Client-facing captures focus on the answer itself and may omit the exact prompt.

Repeated runs are more reliable

Averages, medians and presence rates across repeated runs are materially more dependable than any single answer. Stability is reported alongside position so a single lucky or unlucky run is never mistaken for a trend.

API and consumer interfaces differ

Routine monitoring uses provider APIs; verification uses controlled browser sessions. These can produce different answers. The collection mode is stored with every observation and shown in the evidence viewer.

Raw evidence is never overwritten

Automated interpretations sit on top of immutable raw observations, and human corrections sit on top of those. Every correction records the previous value, the new value, the reason and the reviewer.