How we measure AI visibility

The engine

Both the free check and the rankings ask ChatGPT: OpenAI's model, called through the API with a short instruction to answer like an AI search assistant, name specific options when the question calls for a choice, and cite sources as links.

The model answers from what it learned. It does not run a live web search for these questions, so the results show what the model already believes about a market. The ChatGPT app can search the web for some questions and may answer differently, and answers vary between runs.

Weekly rankings

Each category has ten fixed buyer questions written without brand names. Every week each question is asked once. A second model call lists the companies the answer names; a name only counts if it appears in the answer text, and spelling variants of one company are merged.

Questions are asked in English unless a category says otherwise. These categories are asked in another language, with questions written in that language the way a local buyer types them rather than translated, and the engine is told to answer in it: Japan property sites for Spanish speakers (Spanish, location ES); On-premise AI providers in Spain (Spanish, location ES).

• Named in is the share of answers that named the brand at least once.
• Average position is the brand's order among the brands each answer named, averaged over the answers that named it.
• Named first counts the answers where it was the first brand named.
• Websites are the links inside the answers. From a model without web search, they are sites it remembers, not pages it read.

A week is published only when at least six of the ten questions were answered. Missing answers are left out, never counted as a brand being absent.

The free check

We read the homepage and write buyer questions for that market with one model call, then ask ChatGPT the three with the most buying intent that do not contain the brand name. Each answer scores 100 if the brand is named first, 80 in the top three, 60 in the top five or when its position cannot be read, 40 further down, 20 if only the website is linked, and 0 if missing. The score is the average of the answered questions; an unanswered question is excluded rather than scored as zero.

The crawler table reads the site's robots.txt rules for the homepage. It does not test firewalls or bot challenges.

Changes to the questions

A question is not reworded once it is in use, because that would change what its weekly series measures. New questions and categories are added instead and listed here.

• 17 October 2026: added two categories asked in Spanish, Japan property sites for Spanish speakers and on-premise AI providers in Spain. No existing question changed.

Limits

• Ten questions a week is a sample, not a market census. Small differences between brands are noise.
• One engine is measured. Other assistants, and ChatGPT with web search, can rank brands differently.
• Being named is not being recommended: an answer can name a brand to warn against it.
• Question wording shapes answers. The questions are published on each ranking page so you can judge them.

Questions or corrections: contact UrbanoDX.