What AI Recommends When Diners Ask for Los Angeles Fine Dining
A 50-question audit across Claude, ChatGPT, Gemini and Perplexity reveals a clear top tier, sharp platform divergence and a very different competitive set depending on what the diner is actually trying to decide.
Methodology. Each unbranded buyer question was run three times on Claude, ChatGPT, Gemini and Perplexity, producing 600 usable AI responses. Restaurants are ranked by normalized response appearances: a restaurant counts at most once in each usable individual response, regardless of how many times its name or an approved alias appears.
Entity variants were consolidated before scoring. Examples include Spago → Spago Beverly Hills, Melisse → Mélisse, Ki → Restaurant Ki, Republique → République, 71 Above → 71Above, CUT by Wolfgang Puck → CUT Beverly Hills, CUT → CUT Beverly Hills, Culina Restaurant at Four Seasons → Culina, Culina at Four Seasons → Culina, Mother Wolf LA → Mother Wolf, plus 16 additional approved aliases recorded in the frozen analysis. Closed restaurants, people, publications, hotels, landmarks, platforms and ambiguous non-restaurant entities were excluded from rankings. The analysis uses audit ID la-fine-dining-20260813-201036-f9cf81d3, metric version response_presence_v1, and a 600-response denominator.
Retrieval context: Gemini and Perplexity were web-grounded in all 150 scored responses. ChatGPT used web retrieval in 45 of 150 scored responses and remained model-only in 105. Claude's scored responses did not include retrieval metadata. Separate query-analysis companions were captured for 10 selected prompts and are not counted in the 600 scored responses. This study measures recommendation presence, not reservation conversion or causal impact.
Los Angeles does not have one flat fine-dining leaderboard. Providence and n/naka sit in a tier of their own, appearing in 346 and 295 of 600 responses respectively. After that, the field fragments. Spago Beverly Hills is third overall but becomes the dominant answer for private dining and client entertainment. Hayato and Kato rise when the request becomes tasting-menu specific. Baroo breaks into the top five only when the question is about cuisine and culinary point of view. The buyer situation changes who gets considered.
The same restaurant can look dominant on one platform and nearly absent on another.
Overall rank uses all 600 usable responses. Each platform column has a maximum of 150 responses. Red numbers mark near-zero presence; amber marks a weak platform relative to the rest of the field.
| Rank | Restaurant | Overall / 600 | Claude / 150 | ChatGPT / 150 | Gemini / 150 | Perplexity / 150 |
|---|---|---|---|---|---|---|
| 1 | Providence | 346 57.7% | 106 | 91 | 79 | 70 |
| 2 | n/naka | 295 49.2% | 117 | 77 | 60 | 41 |
| 3 | Spago Beverly Hills | 190 31.7% | 61 | 62 | 44 | 23 |
| 4 | Mélisse | 135 22.5% | 35 | 29 | 31 | 40 |
| 5 | Vespertine | 131 21.8% | 49 | 44 | 20 | 18 |
| 6 | République | 126 21.0% | 48 | 42 | 19 | 17 |
| 7 | Somni | 115 19.2% | 26 | 18 | 36 | 35 |
| 8 | Osteria Mozza | 114 19.0% | 51 | 25 | 22 | 16 |
| 9 | Hayato | 111 18.5% | 29 | 15 | 46 | 21 |
| 10 | Kato | 84 14.0% | 16 | 11 | 33 | 24 |
| 10 | Nobu Malibu | 84 14.0% | 52 | 19 | 10 | 3 |
| 12 | Bestia | 74 12.3% | 30 | 18 | 14 | 12 |
| 13 | 71Above | 60 10.0% | 2 | 10 | 41 | 7 |
| 14 | Citrin | 58 9.7% | 26 | 9 | 14 | 9 |
| 15 | Orsa & Winston | 45 7.5% | 1 | 9 | 25 | 10 |
The top tier is unusually clear. Providence appears in 57.7% of all responses and n/naka in 49.2%; Spago Beverly Hills drops to 31.7%. Platform concentration then becomes a major part of the story: Nobu Malibu is heavily Claude-led, while 71Above and Orsa & Winston rely disproportionately on Gemini.
Visibility risk is often platform-specific, not market-wide.
A restaurant with a healthy overall count can still have a major blind spot on one system. These gaps matter because a consumer's consideration set changes depending on which assistant they use.
| Restaurant | Overall | Strongest Platform | Weakest Platform | Pattern |
|---|---|---|---|---|
| Nobu Malibu | 84 | Claude 52 | Perplexity 3 | Near-zero platform presence |
| 71Above | 60 | Gemini 41 | Claude 2 | Near-zero platform presence |
| Orsa & Winston | 45 | Gemini 25 | Claude 1 | Near-zero platform presence |
| Hayato | 111 | Gemini 46 | Chatgpt 15 | Large platform concentration |
| Kato | 84 | Gemini 33 | Chatgpt 11 | Large platform concentration |
| Spago Beverly Hills | 190 | Chatgpt 62 | Perplexity 23 | Large platform concentration |
The buyer situation changes the competitive set.
Each cluster contains 120 usable responses: 10 buyer questions × 4 platforms × 3 runs.
Prestige & Destination Dining
Destination status, iconic reputation, difficult reservations and luxury-first trip planning.
Providence and n/naka dominate the destination frame. Answers repeatedly position Providence around seafood, tasting-menu polish and reliability; n/naka around modern kaiseki, scarcity and a highly distinctive chef-led experience. Vespertine gains ground when the request rewards singularity and immersion.
Tasting Menus & Culinary Format
Multi-course format, pacing, wine pairings, chef counters and tasting-menu specificity.
Format specificity changes the field. n/naka and Providence are nearly tied, while Hayato and Kato move sharply upward because the answers can describe a precise multi-course format, culinary tradition and chef point of view.
Private Dining & Client Entertainment
Client dinners, business meals, private rooms, executive entertaining and group hosting.
Spago becomes the clear leader here. The answers connect it to Beverly Hills, iconic status, polished hospitality and business entertaining. Redbird also rises because private-event and hosting contexts are more legible than its overall category position suggests.
Special Occasions & Romantic Dining
Anniversaries, proposals, milestone celebrations, romantic dinners and family occasions.
Providence and n/naka remain strongest when the meal needs to feel consequential. Nobu Malibu improves on occasion-led prompts because its oceanfront setting and recognizable LA identity give the models an easy reason to recommend it beyond cuisine alone.
Cuisine & Culinary Point of View
Chef identity, culinary originality, cuisine-specific requests and distinct creative perspective.
n/naka leads when the request is about culinary point of view rather than broad prestige. Baroo jumps into the top five despite sitting outside the overall top 15, with answers repeatedly describing modern Korean tasting menus, fermentation and a more personal narrative.
Four patterns shape the Los Angeles market.
Providence and n/naka are not just first and second. They occupy a different visibility tier.
Providence appears in 346 of 600 responses; n/naka in 295. Both lead two or more high-intent clusters and surface across all four platforms. The drop to Spago at 190 appearances is substantial. AI systems have a highly repeatable story for the top two: Providence as a benchmark seafood tasting-menu destination; n/naka as a coveted, chef-led modern kaiseki experience.
Overall rank can hide the buyer situation a restaurant actually owns.
Spago Beverly Hills ranks third overall, but its strongest commercial position is Private Dining & Client Entertainment, where it appears in 82 of 120 responses. It falls to only 16 appearances in Tasting Menus & Culinary Format and 7 in Cuisine & Culinary Point of View. The same brand can therefore be highly visible for one buying decision and peripheral for another.
The web-grounded platforms materially reshape the middle of the field.
Somni receives 71 of its 115 appearances from Gemini and Perplexity. Hayato receives 67 of 111, and Kato 57 of 84. These are the two platforms that used web grounding in every scored response, making current retrievable evidence especially important to the competitive set they construct.
Grounding helps, but it does not eliminate stale recommendations.
Verified closed restaurant artifacts still appeared in 85 Claude responses, 60 ChatGPT responses, 31 Gemini responses and 9 Perplexity responses. They were removed before rankings were calculated. The finding is methodological but commercially important: current web access does not guarantee current answers, and model knowledge can preserve outdated venues long after closure.
Three signals recur across the restaurants that surface consistently.
A specific, repeatable identity
The strongest restaurants are easy for an AI system to explain in one sentence: seafood tasting menu, modern kaiseki, avant-garde immersion, iconic Beverly Hills institution. Generic prestige is weaker than a precise reason to choose the restaurant.
AI visibility strengthens when the answer can justify the recommendation, not merely name the venue.Evidence tied to the occasion
Spago's private-dining lead and Nobu Malibu's special-occasion lift show that broad brand recognition is not the same as occasion fit. Rooms, views, pacing, group format, service style and the type of experience all give the model a reason to match a restaurant to a specific decision.
Restaurants need enough public evidence for the model to understand when the restaurant is right.Current, retrievable corroboration
Gemini and Perplexity produce a different middle tier than Claude and ChatGPT. Restaurants with strong current editorial and web signals can gain ground quickly, but stale listings and old model knowledge still enter the answer set.
Owned content matters most when it is reinforced by current third-party and local evidence.The opportunity is not to chase a universal score. It is to strengthen the decisions that matter.
Start with the buyer situations where the business wants to be considered. A restaurant trying to grow private dining should not treat a high destination-dining rank as proof that AI systems understand its rooms, capacities, hosting experience or client-entertainment fit. The cluster results show that those are separate recommendation problems.
Then make the restaurant easier to justify. That means clear first-party language around cuisine, tasting format, chef identity, occasion fit, private dining, wine program, views, pacing and other attributes that actually determine a recommendation. The 2024 KDD paper GEO: Generative Engine Optimization found that content interventions could improve source visibility by up to 40% in its experimental setting, while also showing that effective strategies varied by domain. The practical implication is not to mass-produce generic content; it is to make the evidence for selection clearer and more specific to the decision.
Finally, treat platform divergence as a diagnostic signal. A large gap between Claude, ChatGPT, Gemini and Perplexity points to different underlying evidence environments, not one universal visibility problem. Current listings, editorial coverage, structured first-party information and durable brand narratives all matter, but they do not affect every platform in the same way.
The window is open now: restaurants that make their identity, use cases and current evidence easier to retrieve can shape how AI systems describe them before those recommendation patterns harden further.