Methodology
How the AI Recommendation Index is measured, and what it does not say.
Questions
For each category and city we ask 3 wordings, each 2 times per assistant and per week:
- Who is the best {singular} in {city}, {state}?
- Can you recommend a reliable {singular} in {city}, {state}?
- What are the top-rated {plural} in {city}, {state}?
Assistants
Google Gemini (with Google Search), through their official APIs with live web search. Consumer apps may answer differently, for example because of location, history or personalization.
Counting
Each answer is read by a language model that lists the businesses it names (directories such as Yelp or Google are excluded). Variants of the same name are merged. The mention rate is the share of answers naming the business, pooled over the last 4 weekly runs.
Uncertainty
AI answers vary from one run to the next. We report 95% Wilson confidence intervals: if two businesses' intervals overlap, their difference may be noise.
Limits
This measures visibility in AI answers, not quality. Names can be misread or invented by the assistants; open the business's own website before relying on it. Jovian Labs is independent and not affiliated with Google, Perplexity or any business listed.