Every one of the 24 famous New York restaurants we tracked got named by an AI engine at least once. That is the opposite problem to the one you might expect from a city this size: New York restaurants are not being erased from AI answers, they are being rationed. Over 26 days in July 2026 we ran our own daily multi-engine tracking against 35 real dining questions, the kind people actually type, a tasting menu for an anniversary, where to take a client for steak, the best pasta in the West Village, and scored 830 AI answers against 24 of the city's best-known restaurants. The most visible name on the list, Gramercy Tavern, was still only named in 15.1% of those answers. New York's AI dining answer is not built from footfall or fame. It is built from a small, nameable set of sources, and this is the first time we have measured exactly which ones.
TL;DR
- Concentration: Eater was retrieved in 40.1% of New York dining answers, Time Out 36.9%, The Infatuation 35.7%. Add Reddit at 28.4% and the two booking platforms, Resy 25.4% and OpenTable 24.0%, and a handful of sources decide who gets recommended.
- Ceiling: even Gramercy Tavern, the most visible restaurant we tracked, was named in only 15.1% of answers. Fame does not convert into AI presence.
- The Carbone paradox: Carbone ties for only eighth place on visibility (8.2%) but has the most mentions of any restaurant we track (243), the highest share of voice (11.8%) and an average position of 1.8. Visibility and dominance are different metrics.
- Retrieved is not cited: nyc.com is pulled into 8.0% of answers but almost never quoted, a 0.11 citation rate. Reddit is the most heavily quoted source in the whole dataset.
How we measured New York's AI dining answers
We tracked 24 of New York's best-known restaurants, a spread from tasting-menu destinations like Gramercy Tavern and Atomix to institutions like Peter Luger and Keens Steakhouse, against 35 real dining questions run daily across the major AI engines. A 25th tracked entity was our own control brand and is excluded from every restaurant figure below. The window ran from 1 to 26 July 2026 and produced 830 scored answers in total.
Visibility, in this study, means the share of tracked answers in which a restaurant was named. Not every question suits every restaurant, a private dining query and a late-night bar snack query pull different names, so no restaurant should be expected anywhere near 100%, and none of the 24 we track came close.
Three caveats matter more than any single number here. This is a young dataset, a first read, and we will refresh it quarterly. The set is deliberately New York's most famous restaurants, not a sample of the city's roughly 20,000, so what we are measuring is the ceiling, not the average. And visibility is a share of tracked answers, not a verdict on the food.
Which New York restaurants does AI actually name?
Read the top number first, because it sets the scale for everything else. Gramercy Tavern, the most-named restaurant in our tracking, appeared in only 15.1% of answers. Here is the full leaderboard.
| Restaurant | Visibility | Mentions | Share of voice | Avg position |
|---|---|---|---|---|
| Gramercy Tavern | 15.1% | 184 | 8.9% | 3.7 |
| COTE | 12.9% | 162 | 7.9% | 3.3 |
| The Modern | 11.8% | 125 | 6.1% | 3.1 |
| Torrisi | 10.8% | 146 | 7.1% | 4.6 |
| Le Bernardin | 10.8% | 120 | 5.8% | 5.8 |
| Don Angie | 9.9% | 111 | 5.4% | 4.8 |
| Via Carota | 9.2% | 90 | 4.4% | 4.1 |
| The Four Horsemen | 8.2% | 105 | 5.1% | 2.1 |
| Carbone | 8.2% | 243 | 11.8% | 1.8 |
| Atomix | 8.0% | 99 | 4.8% | 5.6 |
| 4 Charles Prime Rib | 7.0% | 69 | 3.3% | 4.5 |
| Semma | 6.5% | 96 | 4.7% | 12.0 |
| Lilia | 6.0% | 76 | 3.7% | 6.4 |
| Dhamaka | 5.3% | 75 | 3.6% | 4.5 |
| Keens Steakhouse | 4.8% | 57 | 2.8% | 1.6 |
| Tatiana | 4.7% | 56 | 2.7% | 10.3 |
| Ruffian | 4.7% | 64 | 3.1% | 3.1 |
| June Wine Bar | 3.4% | 58 | 2.8% | 2.8 |
| Peter Luger | 3.1% | 39 | 1.9% | 3.0 |
| I Sodi | 2.8% | 28 | 1.4% | 6.0 |
| Rezdora | 2.7% | 32 | 1.6% | 7.3 |
| Misi | 1.2% | 14 | 0.7% | 3.0 |
| Union Square Cafe | 0.8% | 7 | 0.3% | 2.0 |
| Parcelle | 0.6% | 7 | 0.3% | 5.0 |
Fame does not track with the table. Peter Luger, arguably the most recognisable steakhouse name in America, sits at 3.1%, below newer, smaller rooms like Don Angie and Via Carota. It also carries the second-lowest sentiment score in the set at 57 out of 100. Being a New York institution buys a restaurant name recognition with diners. It does not buy a place in the sources an AI engine reads.
The ceiling finding: even the most famous room in the city tops out at 15%
Gramercy Tavern is not an obscure choice. It is one of the most decorated dining rooms in New York, and it still missed 85% of the answers we tracked it against. That is the real headline, not its leaderboard position. If the city's most-covered, most-awarded restaurant can only clear 15.1%, the ceiling for AI visibility in New York dining is low for everyone, and most of the headroom under it is unclaimed. We saw a similar shape in London, where the leading group topped out at 12.5% (see the UK Restaurant AI Visibility Index). New York's ceiling sits a little higher, but it is still a ceiling, nowhere near the floor most operators assume their reputation buys them.
The Carbone paradox: fewer answers, total dominance when it appears
Carbone is the clearest illustration in the dataset that visibility and dominance are not the same metric. On raw visibility, Carbone ties for eighth place at 8.2%, a little over half Gramercy Tavern's 15.1%. But look at what happens inside the answers where it does appear. Carbone has 243 mentions, by far the most of any restaurant we track, well clear of Gramercy Tavern's 184 despite appearing in barely half as many answers. Its share of voice is 11.8%, the highest in the set. And its average position is 1.8, meaning that when an engine names Carbone, it tends to name it first or second, not buried in a list.
Contrast that directly with Gramercy Tavern: named in nearly twice as many distinct answers, but occupying less of each answer once it appears, hence the lower share of voice. Carbone gets named less often, and dominates harder every time it shows up.
Position tells a similar story elsewhere in the leaderboard. Keens Steakhouse is only 4.8% visible, one of the lower scores on the table, but it has the best average position of any restaurant we track, 1.6. The Four Horsemen follows at 2.1. Meanwhile Semma and Tatiana are named often enough to matter, 6.5% and 4.7% visibility, but with average positions of 12.0 and 10.3: they are in the answer, buried near the bottom of it, which for a diner skimming a list is close to not being there at all.
Which sources build New York's AI dining answer?
Three editorial titles, one community platform and two booking sites do most of the work of deciding what New York's AI dining answer recommends. Eater was retrieved in 40.1% of the answers we tracked, Time Out in 36.9%, The Infatuation in 35.7%. Add Reddit at 28.4% and the two booking platforms, Resy at 25.4% and OpenTable at 24.0%, and you have covered the sources that shape most of what gets said.
| Domain | Type | Retrieved in | Citation rate |
|---|---|---|---|
| eater.com | Editorial | 40.1% | 1.79 |
| timeout.com | Editorial | 36.9% | 1.22 |
| theinfatuation.com | Editorial | 35.7% | 1.37 |
| reddit.com | Community | 28.4% | 1.85 |
| resy.com | Booking | 25.4% | 1.41 |
| opentable.com | Booking | 24.0% | 0.72 |
| cntraveler.com | Editorial | 18.6% | 1.03 |
| michelin.com | Guide | 16.4% | 1.61 |
| tripadvisor.com | Listings | 12.3% | 0.75 |
| restaurantsforkings.com | Editorial | 9.4% | 2.21 |
| nyc.com | Listings | 8.0% | 0.11 |
| vinepair.com | Editorial (drinks) | 7.4% | 2.03 |
| instagram.com | Social | 6.8% | 1.39 |
| youtube.com | Social | 5.9% | 1.24 |
| theurbanlist.com | Editorial | 5.2% | 1.98 |
| observer.com | Editorial | 4.7% | 2.26 |
| decanter.com | Editorial (drinks) | 4.1% | 0.94 |
| yelp.com | Listings | 3.3% | 1.11 |
| tiktok.com | Social | 2.4% | 1.65 |
Booking platforms carry more weight here than in the London data we published last month. Resy at 25.4% and OpenTable at 24.0% each come close to Reddit's 28.4%, and taken together they are retrieved more often than any single editorial title, which says a New York dining answer is being built partly from availability and reservations, not only from editorial opinion. Drinks-led media is a real route in too: VinePair and Decanter both make the list, and both get quoted hard whenever they are pulled.
Retrieved is not the same as cited
Retrieved means an engine pulled a page from that domain into its working set. Citation rate is how many times, on average, that domain gets quoted inline once it is retrieved, and it can run above 1.0. The gap between the two numbers matters more than either number alone.
nyc.com is retrieved in 8.0% of our tracked answers, a respectable-looking figure on its own, but its citation rate is just 0.11. The engine checks it for a fact and almost never repeats what it says. Compare that with Reddit, retrieved less often overall at 28.4%, but the single most quoted source in the entire dataset: 436 citations at a 1.85 citation rate, behind only Eater on raw citation count. Observer (2.26), restaurantsforkings.com (2.21) and VinePair (2.03) are quoted hardest per retrieval of anything on the list. Chase retrieval alone and you can end up everywhere and quoted nowhere.
How does New York compare with London and Berlin?
We run the same daily tracking method across three cities now, and the shape of the answer changes city by city.
| Top source | Retrieved in | ||
|---|---|---|---|
| New York | Eater (editorial) | 40.1% | 28.4% |
| London | Time Out (editorial) | ~64% of ChatGPT answers | 33% of ChatGPT answers |
| Berlin | visitberlin.de (city tourism board) | 24.9% (44.9% of ChatGPT answers) | 10.1% |
London runs hottest on a single editorial title: Time Out clears roughly two in three ChatGPT answers on its own. Berlin runs coolest and leans on an official city tourism source instead of a restaurant magazine, with Reddit playing a far smaller role than in either English-language city. New York sits in between. No single source dominates the way Time Out does in London, but the concentration is still real: three editorial titles plus Reddit and two booking platforms cover most of it.
What does a broader study say about the rest of the city?
Our 24 restaurants are New York's most famous, so they tell you about the ceiling, not about the roughly 20,000 restaurants in the city we did not measure. For that broader picture, Local Falcon's Restaurant AI Visibility Index, a separate study that ran 10,000 US restaurants through Google Maps and Google AI Overviews across nine grid points per location in May 2026, found 90.4% of the 115 New York restaurants it sampled invisible at every grid point checked, the highest invisibility rate of the 30 cities in its sample. That is a different method to ours, local grid search rather than dining-question answers, and a much broader and more representative sample than our 24 famous names. We cite it here, clearly separated from our own tracking, as a reminder that the leaderboard above is the top of the city's AI visibility, not the middle of it.
What this means if you're opening a restaurant in New York
UK and European restaurant groups that open a New York room start from zero in this data: no Eater coverage, no Reddit threads, no Resy history, nothing an AI engine has learned to associate with the name yet. The source list above is the map for that cold start, the specific outlets to brief, the community threads to be honestly present in, the booking platforms to keep current from day one, rather than a generic PR plan. We wrote a fuller playbook for exactly this problem in getting AI-visible when your restaurant expands to a new city.
Disclosure: Schmitdy is our own AI search service at AI Heroes, and this study comes from the same daily multi-engine tracking we run for the restaurants we work with. We publish it openly because a New York read did not exist yet, and we will refresh it quarterly. For how this concentration pattern compares with what we found in London, see which sources ChatGPT cites for London dining and Reddit and local forums behind restaurant AI answers. If you want to see where your own restaurant or group sits against this source list, the free AI search audit shows it.
