Original research · November 1, 2026
How AI actually picks restaurants
We collected 650 answers from ChatGPT and Claude about where to eat around Washington, DC — with live web search on — and recorded every source they read before answering. That came to 4,159 citations across 65 restaurants.
The short version
Two websites account for 69.8% of everything AI reads before recommending a restaurant. Neither of them is the restaurant's own website.
Where the answers come from
Every source cited across 650 answers, counted by domain.
It's not Google search — it's Maps
Splitting the two biggest sources by the kind of page cited changes what the finding means. AI isn't reading search results pages about your restaurant. It's reading your Google Business Profile, and Yelp's ranked category pages.
| Page type | Citations | Share |
|---|---|---|
| Google Maps / Business Profile | 1,904 | 45.8% |
| Yelp search results | 962 | 23.1% |
| Yelp business listing | 30 | 0.7% |
| Yelp (other) | 7 | 0.2% |
Does review count predict getting recommended?
Each restaurant's Google review count against how many of 10 AI answers named it, across the 65 restaurants we could match to a Google Business Profile.
| Google reviews | Restaurants | Named in |
|---|---|---|
| Under 100 | 2 | 2 of 10 |
| 100–499 | 12 | 3.9 of 10 |
| 500–1,499 | 32 | 4.3 of 10 |
| 1,500+ | 19 | 7.1 of 10 |
Correlation between review count and AI visibility score across 65 restaurants: r = 0.38. That is a moderate relationship: review volume helps, but it is clearly not the only thing going on.
What this means if you run a restaurant
- Your website is almost irrelevant here. Restaurant domains barely register in the citations. Redesigning it will not move this.
- Your Google Business Profile is the single biggest lever. It is the most-read source by a wide margin, and it is the one you directly control.
- Yelp still matters more than people think. Not your listing — your position in Yelp's category results for your area.
- Local "best of" lists are the long tail. Publications like The Infatuation and local magazines make up a real slice. Getting into one is worth more than a hundred directory listings.
Method
For each restaurant we generated five questions a customer in its area might ask when choosing that kind of place, then put each to both ChatGPT and Claude with live web search enabled and location set to the restaurant's own town — ten answers per restaurant. We recorded whether the restaurant was named, which competitors were named instead, and every URL each model cited. Scores are simply the share of those ten answers that named the business. AI answers change over time; this is a snapshot, not a permanent ranking. Restaurants were drawn from a DC-area directory and are not a random sample of the region.
See your own numbers
Run the same questions against your business, free, and see whether AI names you or a competitor.