Method · AI visibility

AI visibility is a sample, not a fixed ranking.

An answer changes with the question, language, location, model and date. A useful report preserves those conditions and the sources behind each observation.

Start with a buying decision

For example: “Which independent high-jewellery boutiques in Geneva should I shortlist for an engagement ring?” Keep branded questions separate from category discovery. A model repeating a brand supplied in the prompt is not evidence of discovery.

Record a reproducible protocol

Freeze the prompt set, market, language, date, model or product version, search setting and session conditions before collection.
Repeat each prompt in independent sessions. Define the number of repetitions in advance; retain every run, including failures and refusals.
Preserve the answer and its cited URLs. A mention, recommendation and citation are different observations.

Use denominators people can inspect

Mention rate: Valid answers naming the brand ÷ all valid answers in the stated sample.
Citation rate: Valid answers citing an owned canonical URL ÷ all valid answers in the same sample.
Evidence quality: Separate direct observation from interpretation; report missing responses and variability across repeated runs.

Do not turn visibility into a sales claim

These metrics describe sampled answers. They do not prove revenue, future recommendations or causality. A before-and-after comparison needs comparable conditions and a record of model and search changes.

European luxury observatory: protocol first

A 50-brand study is a proposed research scope, not a published result. No league table or benchmark is presented until a dated sample and auditable evidence file exist.