How we measure the Index

What we do

• For each category we write buying questions in the phrasing a person actually uses, not brand names.
• We ask each question of each model and record the full answer.
• We extract brands named in every answer and publish only their share of answers.
• Every category page prints prompts, models, sample size and date so its measurement can be reproduced.

What this does not mean

Being named often is not being good. Assistants name what they have read about, which can favour brands that are excellent, old, loud, or simply written about more. The Index measures AI visibility and nothing else.

We do not claim the figures are stable. Models and indexes change, and the same question may get a different answer next month. That is why we measure repeatedly.

Where the data comes from

Our own runs, on our own account, paid for by us. Nothing on these pages is derived from any customer’s data — not their prompts, not their tracked competitors, not their results. That boundary is enforced in the code that generates these pages, not just by convention.

Corrections

If a figure looks wrong, tell us. We will re-run the category and correct or remove the page. A measurement we cannot stand behind is worse than no page.