# The AI Restaurant Recommendations Report: United States

A Searchable data study of which restaurants AI chatbots recommend in 55 US cities, and how much of each city's highest-rated restaurants they leave out. Data correct as of October 2026.

## Methodology

- Searchable asked ChatGPT, Google Gemini, Google AI Mode and Perplexity for restaurant recommendations in 55 major US cities and tourist destinations over six days in October 2026, through their standard consumer interfaces. This produced 174,142 answers.
- Each city got 20 questions: the main "where to eat" question in three wordings, a long-list request, hidden gems, occasions, independent restaurants and nine types of cuisine.
- Each city's comparison list is its highest-rated restaurants: those at the top of Tripadvisor's ranking for the city, rated 4 stars or more with at least 50 reviews. Every listed restaurant was checked on Google Maps by name, address and phone number; closed restaurants and non-restaurants were removed, as were restaurants that both Tripadvisor and the legal city boundary place in another town, and each was replaced by the next-ranked restaurant on the list. That left between 52 and 98 restaurants per city. Because the lists differ in length, one city shouldn't be compared with another on the share of its highest-rated restaurants that AI recommends or never names.
- Restaurants named in the answers were counted under every name the chatbots used, whether or not they are on Tripadvisor's list. Every off-list name that a published figure depends on was checked to confirm it is a real restaurant, and headings, dishes, neighborhoods and names that could not be found anywhere were removed.
- "Regularly recommended" means named in at least 10% of answers to the main questions, with each chatbot weighted equally. "Never named" means no answer to any general question names the restaurant.
- "Repetition" means the share of answers naming a question's five most-named restaurants, per chatbot, then averaged.

## Key findings

1. AI regularly recommends just 1 in 8 (11.7%) of each city's highest-rated restaurants on Tripadvisor, and never names 1 in 12 (8.4%).
2. A city's five most-recommended restaurants appear in 56.6% of answers to the main "where to eat" questions. Tampa is the most repetitive (70.9%), followed by Santa Fe and Key West. San Diego is the least (41.1%).
3. Asking again rarely brings something new: a third of repeat answers (32.5%) name a restaurant that no earlier answer to the same question had. In Anchorage only 21.3% do. New York City, Las Vegas and San Francisco give the most variety.
4. In many cities one restaurant is close to a guaranteed mention: Tampa's Columbia Restaurant is named in 85.8% of answers about where to eat there, Savannah's Olde Pink House in 84.9% and Indianapolis's St. Elmo Steak House in 84.2%.
5. In New York City, AI regularly recommends only 5% of the city's highest-rated restaurants on Tripadvisor.
6. Asking for a cuisine does not widen the answer. Ask for a steakhouse and the same five come back in 73.5% of answers; every cuisine repeats its top five in more than half of answers.
7. ChatGPT names about twice as many restaurants per answer as Google AI Mode: about 11 for ChatGPT, 6 for Gemini, 5 for Google AI Mode and 2 for Perplexity. Their top 10 lists for the same question and city overlap by only 30%.
8. Eater (34.8%) and Tripadvisor (34.5%) are the websites AI cites most when it recommends a restaurant. Google AI Mode leans on Google's own business listings more than on any outside site.
9. All four chatbots favor restaurants with stronger Google reviews. Restaurants rated 4.7 stars or higher are named more than four times as often (8.2% of answers) as those rated under 4.5 (1.9%), and those with more than 4,000 reviews over five times as often (7.0%) as those with fewer than 500 (1.2%).

The restaurants named are not Searchable customers and were not involved in this analysis.