AI Visibility Index

Published editions

What has been measured

Separately, 8 questions are repeats, asked twice to measure how much the answers move. They are excluded from every brand's share, so the figures below cover the 100 distinct questions only.

Each question is put to every engine separately, so 100 questions across 2 engines is 200 attempted answers. Of those, 191 came back usable and are what each brand's share is measured against; the rest failed or returned no AI answer and are excluded rather than counted as a brand not being cited.

What this means if it is your brand

Two useful things follow from the numbers above, and Robot Visible was built around both.

How each edition is produced

Questions come first, brands never appear in them

Each category uses around fifty non-branded buyer questions covering discovery, comparison, use-case and problem intents. Naming a brand in a question would prompt its own answer.

Every brand domain is verified before a run

Saying a brand was not cited is a claim about a real company. Domains are checked against the company's own site first, because attributing the wrong domain would make that claim false.

Bands, not a numbered ranking

Fifty questions cannot support ordinal precision. Brands are grouped by how often they were cited, and adjacent brands are not presented as beating one another.

Each edition publishes its own error bar

Several questions are asked twice in the same run and the difference is published. Stability varies by category: a consolidated market barely moves, a fragmented one churns at the margins.

Inputs are frozen when a run starts

Questions, brands, thresholds and engines are snapshotted, so editing a category later can never move a past month's figures.

Our wider rules for publishing figures about other companies, and a dated log of every correction we have made, are on the methodology and corrections page.