Ally Kiel Consulting · AI Visibility Research · July 2026
AI Recommendation Patterns in New York City Fine Dining
A 1,199-response audit of how Claude, ChatGPT, Gemini and Perplexity construct the city’s fine-dining shortlist—and which restaurants enter the recommendation set.
Category Fine-dining restaurantsMarket New York CityAudit date July 30, 2026
50Unbranded diner prompts
4AI platforms
5Query clusters
3×Runs per query form
1,199Usable responses
Methodology
Measuring recommendation inclusion, not restaurant recognition
The audit asked unbranded questions that a prospective diner, event planner or host might use when choosing a fine-dining restaurant. It did not ask the platforms what they knew about any named restaurant.
How to read the results
Fifty prompts were organized across five booking-intent clusters. Each prompt was run three times in its original form and three times with controlled query variation across four AI platforms, producing 1,200 expected trials. One Gemini response in Prestige & Destination Dining returned an explicit provider error, leaving 1,199 usable responses and 99.92% coverage.
Primary metric: normalized response appearances. A restaurant counts no more than once per response, even if the answer repeats its name. Overall inclusion rate is appearances divided by 1,199 usable responses. Platform denominators are 300, except Gemini at 299.
Comparison set: the extraction layer produced 1,156 strings. The published analysis uses a manually reviewed set of 133 current, eligible restaurant entities covering every name with enough frequency to affect the leaders or concentration cases.
Entity normalization: material consolidations included Le Bernardin Privé / Le Bernardin Prive → Le Bernardin; Restaurant Daniel / Daniel by Daniel Boulud / The Skybox at Restaurant Daniel → Daniel; The Modern at MoMA / The Modern – Dining Room / The Modern - Dining Room / The Bar Room at The Modern / The Kitchen Table at The Modern → The Modern; Chef's Counter at Atomix → Atomix; The Dining Room at Gramercy Tavern → Gramercy Tavern; Brooklyn Fare / The Restaurant at Brooklyn Fare → Chef's Table at Brooklyn Fare; Al Fiori / Aifiori → Ai Fiori; The Grill at Seagram Building → The Grill; Jungsik New York → Jungsik; Cote NYC / COTE Korean Steakhouse / COTE 550 → COTE; One if by Land → One if by Land, Two if by Sea; Noz / Noz on Park → Sushi Noz; Hawksmoor NYC / Hawksmoor New York → Hawksmoor; Oceana's Chef's Counter → Oceana; Musket Room → The Musket Room; Nobu Downtown / Nobu Midtown / Nobu Tribeca / Nobu Fifty Seven / Nobu 57 → Nobu; Keens → Keens Steakhouse; Sixty Three Clinton → 63 Clinton; Tatiana by Kwame Onwuachi / Tatiana by Chef Kwame / Tatiana by Chef Kwame Onwuachi → Tatiana; Odo East Village → odo; Sushi Yoshino / Yoshino New York / Yoshino • New York → Yoshino; HAGS East Village → HAGS; Polo Bar → The Polo Bar; Le Pavillon NYC → Le Pavillon; Nakazawa / Sushi Nakazawa NY → Sushi Nakazawa; Gage and Tollner → Gage & Tollner; Milos Hudson Yards → Estiatorio Milos; L'Artusi Supper Club → L'Artusi; Shion / Shion 69 Leonard / Shion 69 Leonard St / Shoji at 69 Leonard / Shoji at 69 Leonard Street → Shion 69 Leonard Street; L'abeille à Côté → L'Abeille; River Café → The River Café. Restaurants under common ownership were not combined because diners encounter them as distinct booking choices.
Disambiguation: chef references were not counted as restaurant appearances. “Daniel Boulud” did not count as Daniel, and “by Jean-Georges” did not count as the Jean-Georges flagship unless the restaurant itself was also named.
Exclusions: people, hotels, landmarks, booking platforms, publications and event venues were removed. Examples include Central Park, MoMA, Eric Ripert, The Langham, Tock and Eater NY. Closed or out-of-market restaurants—including Momofuku Ko, Del Posto, Aureole, Kajitsu, 21 Club and Blue Hill at Stone Barns—were also ineligible. Current-status review used official restaurant sources and the Michelin Guide.1
Scope: an appearance measures inclusion, not sentiment, placement, factual accuracy or purchase conversion. The results are a dated observational benchmark, not a permanent model ranking.
Executive read
One citywide default, followed by a context-dependent field
Le Bernardin appeared in 791 of 1,199 usable responses. It ranks first on every AI platform and leads all five buyer-intent clusters—from destination dining and tasting menus to private events and special occasions.
Le Bernardin is not winning one version of fine dining.
It remains legible as a bucket-list restaurant, a tasting-menu destination, a client-entertaining venue, a special-occasion choice and a restaurant with a distinct culinary identity. Every other restaurant depends more heavily on a particular decision context or platform.
66.0%Overall inclusion rate
Platform divergence
The overall ranking hides four different recommendation systems
A restaurant can look established in the combined results while depending heavily on one platform. Counts below show how many responses on each platform included the restaurant.
#
Restaurant
Claude
ChatGPT
Gemini
Perplexity
Total
Inclusion
1
Le Bernardin
253
213
159
166
791
66.0%
2
Eleven Madison Park
215
189
92
135
631
52.6%
3
Daniel
207
176
103
104
590
49.2%
4
Per Se
182
173
63
106
524
43.7%
5
The Modern
100
106
116
95
417
34.8%
6
Gramercy Tavern
134
63
138
60
395
32.9%
7
Atomix
118
59
89
80
346
28.9%
8
Masa
105
49
45
46
245
20.4%
9
Jean-Georges
32
80
77
54
243
20.3%
10
Gabriel Kreuther
81
44
37
29
191
15.9%
11
Chef's Table at Brooklyn Fare
69
42
31
43
185
15.4%
12
The River Café
39
31
41
24
135
11.3%
13
Carbone
43
25
19
40
127
10.6%
14
Ai Fiori
8
8
84
20
120
10.0%
15
Aquavit
55
8
25
22
110
9.2%
Red indicates zero appearances; amber indicates five or fewer. The total is out of 1,199 usable response trials. Claude, ChatGPT and Perplexity are each out of 300; Gemini is out of 299.
Claude7.67eligible restaurants per answer
100.0% of responses named at least one eligible restaurant
ChatGPT6.13eligible restaurants per answer
99.3% of responses named at least one eligible restaurant
Gemini7.63eligible restaurants per answer
99.3% of responses named at least one eligible restaurant
Perplexity5.32eligible restaurants per answer
91.0% of responses named at least one eligible restaurant
These density figures use the 133-restaurant normalized comparison set, not every restaurant name in the raw answers. Gemini used web grounding in 299 usable trials. ChatGPT used web retrieval in 97 of 300, with retrieval most common in Special Occasions and Cuisine & Culinary Point of View. Comparable retrieval metadata was not available for Claude or Perplexity.
Concentration risk
A visible restaurant can still depend on one model
These restaurants illustrate strong platform dependencies. “Largest-platform share” is the portion of a restaurant’s total appearances supplied by a single AI platform.
Restaurant
Strongest platform
Weakest platform
Largest-platform share
Total appearances
Ai Fiori
Gemini (84)
Claude (8)
70.0%
120
Le Coucou
Gemini (48)
Claude (1)
64.0%
75
Don Angie
Claude (50)
ChatGPT (4)
68.5%
73
Essential by Christophe
Gemini (37)
Claude (0)
72.5%
51
Hawksmoor
Gemini (34)
Claude (0)
94.4%
36
Yoshino
Claude (21)
ChatGPT (0)
61.8%
34
Ai Fiori is the clearest high-ranking concentration case: 84 of its 120 appearances come from Gemini. Essential by Christophe and Hawksmoor show an even narrower pattern, while Don Angie and Yoshino lean toward Claude. These are different visibility problems despite similar combined totals.
Results by cluster
Different booking needs produce different shortlists
Four clusters contain 240 usable trials each. Prestige & Destination Dining contains 239 because of the single Gemini provider error.
Bucket-list meals, established authority, hospitality and destination appeal
1
Le Bernardin
212/2392
Eleven Madison Park
185/2393
Per Se
176/2394
Daniel
174/2395
Atomix
125/239
Le Bernardin leads decisively, followed by Eleven Madison Park, Per Se and Daniel. Atomix moves into fifth, showing that contemporary tasting-menu authority can enter the destination set even while the broadest prestige language still favors established institutions.
02
Tasting Menus & Culinary Format
Chef’s counters, menu length, wine pairings, formats and price points
1
Le Bernardin
131/2402
Eleven Madison Park
108/2403
Per Se
95/2404
Atomix
94/2405
The Modern
92/240
Le Bernardin remains first, but the format-specific set narrows. Atomix rises to fourth and Chef’s Table at Brooklyn Fare to seventh, while Gramercy Tavern and The Modern benefit from offering multiple ways to experience the restaurant rather than one rigid format.
03
Private Dining & Client Entertainment
Executive dinners, board meetings, private rooms, buyouts and discretion
1
Le Bernardin
179/2402
Eleven Madison Park
141/2403
Daniel
139/2404
Gramercy Tavern
126/2405
Per Se
111/240
This cluster rewards explicit operational evidence. Gramercy Tavern rises to fourth, The Grill to seventh and Ai Fiori to eighth. The River Café reaches ninth through repeated associations with views, private events and group occasions—attributes that generic prestige alone cannot supply.
04
Special Occasions & Romantic Dining
Proposals, anniversaries, weddings, birthdays and family milestones
1
Le Bernardin
152/2402
Daniel
121/2403
Eleven Madison Park
113/2404
Per Se
92/2405
The Modern
77/240
The River Café rises to seventh and One if by Land, Two if by Sea to eighth because AI systems connect atmosphere, views and romance to a specific decision moment. Le Coucou enters the top ten here despite ranking lower overall.
05
Cuisine & Culinary Point of View
Cuisine, innovation, wine, vegetable-forward dining and culinary identity
1
Le Bernardin
117/2402
Eleven Madison Park
84/2403
Atomix
75/2404
Daniel
65/2405
Jean-Georges
52/240
Atomix moves to third and Jean-Georges to fifth. Marea enters seventh through seafood and Italian associations, while Masa remains prominent through Japanese tasting-menu authority. This is the cluster where a distinct culinary identity most clearly competes with general institutional fame.
Key findings
What the city’s recommendation structure reveals
Finding 01
Le Bernardin is the only full-spectrum default
It leads all five clusters and every platform. Its 791 appearances exceed second-place Eleven Madison Park by 160 and fourth-place Per Se by 267. That lead is not tied to one culinary format or one kind of diner; the restaurant is repeatedly treated as the safe answer to materially different high-stakes dining decisions.
Finding 02
The Modern has the most balanced top-tier footprint
Its platform counts—100 on Claude, 106 on ChatGPT, 116 on Gemini and 95 on Perplexity—are unusually even. Gramercy Tavern ranks one place lower overall but reaches the same neighborhood through a very different pattern: strong Claude and Gemini visibility, with much lower ChatGPT and Perplexity inclusion.
Finding 03
Operational context can outrank general prestige
The River Café ranks twelfth overall but seventh for special occasions and ninth for private dining. Gramercy Tavern rises to fourth for client entertainment. Ai Fiori reaches eighth there despite ranking fourteenth overall. When the diner specifies privacy, group format, atmosphere or a milestone, explicit experience evidence reshapes the shortlist.
Finding 04
AI’s restaurant memory is not the same as current availability
The raw extraction repeatedly surfaced closed restaurants and historical artifacts, including Momofuku Ko, Del Posto, Aureole, Kajitsu and 21 Club. They were removed from the eligible rankings. A restaurant category audit therefore requires current-status validation; otherwise stale institutional fame can be mistaken for bookable visibility.
What drives AI visibility
Three signals separate durable authority from isolated relevance
Signal 01
Multi-context authority. Restaurants become defaults when the evidence connects them to more than cuisine or awards: service, atmosphere, occasion, format, private dining and a clear point of view.One strong association can win a cluster. It does not create citywide authority.
Signal 02
Decision-specific operational evidence. Private-room capacity, buyout options, menu formats, timing, wine pairings, accessibility and occasion suitability give AI systems a reason to choose one restaurant for a specific request.Reputation creates consideration. Operational specificity converts it into relevance.
Signal 03
Current, cross-platform reinforcement. Trained-model memory, live retrieval and source selection do not reward the same evidence equally. A restaurant concentrated on one platform—or remembered after it closes—does not have durable recommendation authority.The goal is repeated, current relevance across different recommendation environments.
What restaurants can do with this
Build the evidence for the dining decisions you intend to own
Turn hospitality into answerable evidence. “Exceptional service” is too broad to support a specific recommendation. A restaurant should explain the experience in concrete terms: dining formats, room options, group sizes, timing, dietary flexibility, wine support, privacy and what makes a particular occasion work there.
Publish for retrieval, not keyword repetition. KDD 2024 research found that credible citations, relevant quotations and statistics improved source visibility in generative-engine responses; keyword stuffing did not.2 For restaurants, that means pairing a cohesive owned narrative with current menus, precise private-dining details, attributable recognition and structured factual information.
Close the actual platform gap. Ai Fiori does not have the same problem as Don Angie, and Essential by Christophe does not have the same problem as The Modern. The corrective strategy should follow the platform and buyer-intent pattern rather than applying one generic “AI optimization” plan to every restaurant.
AI recommendation systems are already constructing a New York fine-dining shortlist before a diner reaches a reservation page. The window is open now for restaurants to participate in the evidence those systems use.
Michelin Guide, “New York City Restaurants,” current directory accessed July 2026. Michelin Guide New York. Restaurant operating status was also checked against current official restaurant pages where needed.
Pranjal Aggarwal et al., “GEO: Generative Engine Optimization,” Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2024. https://doi.org/10.1145/3637528.3671900.