- The questions
- They are written for each location in a diner’s words. Most give a neighbourhood, an intersection or a postal code and do not name the restaurant.
- The three assistants
- Each question goes to ChatGPT, Gemini and Perplexity with web search switched on. Every answer is saved word for word with its date.
- Whether the restaurant is named
- We record whether the location appears in each answer and where it sits in the list. An assistant sometimes asks a question back, and those replies are counted separately.
- Which restaurants are named instead
- We list the competing restaurants the assistants name, with the number of answers each one appears in.
- Hours, menu and location
- We read what each assistant says about the restaurant and mark each detail that is wrong or out of date.
- The websites each answer cites
- We record the websites each answer links to, including review websites and delivery platforms. We note whether the restaurant’s own website is one of them.
- Each location’s own sources
- We then check the website, the menu, the Google Business Profile, the reviews and the directory listings for the location against what the assistants said.
- What was not checked
- Some review and social websites block automated reading. The report lists those as not collected and does not estimate what they say.