AI visibility benchmarks by industry

Below 5 brands we publish no percentiles, only the sample size. That floor is enforced in the database (migration 0012, c_min_n=5 — the percentile columns are stored as NULL below it), not as a front-end toggle. With only one or two brands in an industry, publishing percentiles would effectively publish those brands' actual scores.

Each brand is sampled daily across a set of Chinese and English prompts on each AI engine; we record whether it is mentioned and cited. The full method — including how citations are judged and how thin samples are flagged — is published in full on the methodology page and is reproducible.

No. Two clauses of our published white-hat position cover this directly: we don't inflate volume (no re-sending the same prompt to lift a mention rate) and we don't manipulate training data — OrcaScope is a monitoring tool, not an “AI reputation laundering” service. The score only reflects how the engines actually answer; there is no dial on our side to turn.

We don't disclose that, and won't. Only percentiles and sample sizes are published — no brand names, no raw answers, no rankings. That isn't caution: if the aggregate could be reversed back to an individual customer, this page shouldn't exist.

Not necessarily. A percentile tells you where you rank inside the industry. If the whole industry is rarely mentioned by AI, even P90 can be a low absolute score. Read the percentile alongside the raw number.

It rolls forward with the nightly run; each page shows the most recent date that industry actually produced an aggregate. If a run fails, the date stays where it was — we don't relabel yesterday's data as today's.