How RecoRadar measures AI recommendations
RecoRadar shows how often ChatGPT, Claude and Gemini recommend a local business when someone asks for that kind of business in that city, and who they recommend instead.
Updated October 11, 2026
- 6
- customer questions, written for each business
- 3
- AIs: ChatGPT, Claude and Gemini
- 18
- answers in every audit
What we ask
For each audit we write six questions a real customer might type, such as "best dentist in Austin" or "emergency dentist in Austin open Saturday". You can edit them before the audit runs. The questions never include the business's name.
How we ask
We send each question to the official APIs of OpenAI (ChatGPT), Anthropic (Claude) and Google (Gemini) with web search turned on, the way the consumer apps answer. Where an API supports it, we pass an approximate location for the city; the city is always in the question too. We never scrape chatgpt.com, claude.ai, gemini.google.com or Google results.
What we measure
From each answer we record which businesses it recommends and in what order, and which websites it cites. A business counts as named when its name matches closely (allowing for small spelling differences) or when the answer cites its own website. The score is the share of answers that named the business.
How to read the results
AI answers vary from one run to the next and from one person to another. Treat the results as likelihoods, not fixed rankings. The fix list only uses what we measured: which answers named whom and which sites they cited. We don't claim ratings, review counts or rankings we did not fetch.