Visibility Index

How we measure

We ask ChatGPT the questions buyers ask, three times each, in a fresh logged-out browser with no history. We record every brand it names and every source it cites, and report how often each appears. One answer is noise. Three show the pattern.

Checked Sept 23, 2026

What we do, step by step

  1. Pick questions a real buyer would type, in plain words, for each category. We publish every question on the category page.
  2. Open chatgpt.com in a new browser with no account, no memory and no history, using ChatGPT's default settings.
  3. Ask the question and wait for the full answer.
  4. Save the answer text and every source ChatGPT attaches, with its URL, title and publisher.
  5. Close the browser. Repeat twice more, fresh each time.
  6. Read each answer and list the brands it names, in order.
  7. Report each brand's share: in how many of the answers it appeared.

Why three runs

ChatGPT does not give the same answer twice. In our own tests, the top few brands repeat almost every time while the rest change from run to run. A single answer can make a brand look like it is winning or missing when it is neither. Three runs separate the steady names from the noise.

Why the real app, not the API

Many visibility tools ask ChatGPT through OpenAI's developer API. That is cheaper and simpler, but it is not what people see. Published tests have found only about a quarter of the top brands overlap between the API and the logged-in app for the same question. We use the consumer app so our results match what a buyer sees.

What this cannot tell you

  • Logged-in users get personalized answers. Memory and past chats change results. We measure the neutral, logged-out answer.
  • Settings change answers. Users who pick a slower, deeper thinking mode get different sources. We use the default.
  • Location matters. We ask from the United States.
  • Ads. Ads show mainly to logged-in Free and Go users, so logged-out runs rarely see them. We record any we see and will add logged-in checks.
  • It is a sample. Three runs of two questions is a snapshot for one week, not a law.

Independence

We are not affiliated with OpenAI or any brand we measure. Brands cannot pay to change a result. If we ever earn money from a link on a page, the page will say so.