1. Write the questions buyers ask
Start with 15 to 30 questions in the buyer's words, mixed across intents: "best X in Y", "how do I …", "X vs Y" and "alternatives to Y". Keep your brand name out of them; a question that names you measures recall, not discovery. Questions about you by name ("is X legit?") are worth asking too, but only to check the answer is accurate.
2. Ask from the right place
Answers change with location. Ask as a searcher in your market: a city for local businesses, the country for online ones. Engines accept location differently: ChatGPT takes a country, Google's AI Overviews and AI Mode take coordinates, Perplexity takes a country code. Record the location each answer actually used.
3. Ask more than once
AI answers vary between runs. In our own measurements, two same-day runs of the same question on Google's AI Overviews shared only 24% of their cited sources (9 pairs). Use at least 5 runs for a baseline and 2 to 3 a week after that. A single run is a snapshot, not a measurement.
4. Count mentions and citations separately
| Metric | Counts |
|---|---|
| Mention rate | Answers that name your brand ÷ answers |
| Citation rate | Answers that cite a page on your domain ÷ answers |
| Share of voice | Your mentions ÷ all tracked brands' mentions |
| First-mention rank | Your position among the brands an answer names |
Match brands carefully: by domain in the cited links first, then by the engine's own brand cards, then by name on word boundaries. Common-word names (Sage, Apex, Wave) need a check that the word means the company.
5. Report numbers with their sample size
"Named in 40%" means little without n. Show "6 of 15 answers" and a confidence interval; with 15 answers the 95% interval around 40% runs from about 20% to 64%. Call a change only when a significance test says it is real.
6. Record what the AI searched
ChatGPT exposed its web searches in 100% of the 25 answers we measured. Those searches tell you what to rank for. Most trackers skip them.
Doing it by hand, or not
Five engines, 25 questions and 3 runs is 375 answers a week. Our AI visibility tracker runs this protocol for you from $29 a month; the free check runs one snapshot now.
Interval: Wilson score interval, 95%. Measurements: Sightings's own runs through DataForSEO, October 2026.