How we measure AI visibility

When someone asks an AI assistant who to call for a broken AC, a leaking pipe or a toothache, it names a few local businesses. We measure which ones, how often, and in what order. This page explains exactly how, including what the numbers can and cannot tell you.

1. The questions

For each city and industry we ask five questions written the way customers actually ask, from “What is the best HVAC company in Tampa, Florida?” to “My AC stopped working. Who should I call?”. Every ranking page lists the exact questions used.

2. How we ask

• We use OpenAI's API (model gpt-5-nano) with web search turned on and the city set as the searcher's approximate location. This is close to, but not identical to, what someone sees in the ChatGPT app. Every result shows the model and the date it was measured.
• AI answers change from one conversation to the next, so we ask each question several times. A ranking based on five questions asked three times each rests on fifteen answers.

3. Reading the answers

A language model lists every business named in each answer, in order. We keep only names that appear word for word in the answer, skip directories such as Yelp or Angi, and merge spelling variants of the same company (for example “Del-Air Tampa” and “Del-Air Heating & Air Conditioning”).

4. Scores

• Named in X of Y answers is the plain count, and the main number we show.
• Average position is where a company appears in the answers that name it (1 means first).
• Visibility score (0–100, in the free check and reports) weights early mentions more: positions 1–3 count fully, 4–5 count 70%, 6 and later 40%, and an answer that doesn't name the business counts 0. The score is 100 × the average over all answers. Named first in every answer is 100; never named is 0. Competitors are scored the same way, from the same answers.
• Being cited as a source (a link to your website in the answer) without being named doesn't count as a recommendation. We show it separately.

A measurement needs at least 13 of its 15 answers to come back; otherwise it isn't published or used and is tried again later. Free checks use the newest measurement of the city and say how old it is; past 45 days the result says so, and a city people keep asking about is measured again. Paying customers' cities are measured every week.

5. Limits

• Small samples move. Between two measurements a day apart, a company can go from being named in 1 of 15 answers to 4 of 15. Look at the pattern over several weeks, not one number.
• We currently measure ChatGPT. Other assistants such as Gemini and Perplexity can recommend different companies.
• Rankings record what the AI said. They are not our endorsement or a review of service quality.

Corrections

If your business was merged with another by mistake, or you spot an error, email hello@bemention.com and we will fix it.