What the
Machines
Said

Providence leads the Los Angeles fine dining dataset with 346 normalized response appearances — a 57.7% inclusion rate across 600 scored responses. Claude leads the platform count at 106, followed by ChatGPT at 91, Gemini at 79, and Perplexity at 70. All four machines agree. But the margin separating Providence from n/naka in second is only 51 appearances. Compare that to New York, where Le Bernardin leads Eleven Madison Park by 160. Los Angeles has a consensus leader, but the top of the map is more contested — and the platforms diverge significantly below it. The combined appearances of the top four (Providence, n/naka, Spago Beverly Hills, Mélisse) is 966 across 600 responses.

n/naka holds the second spot in the dataset with 295 total normalized response appearances. But the platform profile is not like Providence's. Claude accounts for 117 of those appearances — the single highest platform count for any restaurant in this dataset. ChatGPT has 77. Gemini has 60. Perplexity has 41. n/naka is the restaurant where the gap between Claude and every other platform is widest. Niki Nakayama's modern kaiseki is built for exactly the kind of intensive culinary writing that becomes training data: 18-course menus, long critical analyses of ingredient sourcing, documentary coverage, years of food-press attention. Claude absorbed it. The other three platforms didn't absorb it nearly as fully.

The pattern extends beyond n/naka. Claude leads Osteria Mozza at 51 (more than double ChatGPT's 25), Nobu Malibu at 52 (ChatGPT 19, Perplexity 3), and Vespertine at 49. The restaurants Claude recommends most heavily are the restaurants that culinary writing has documented most intensively over time. That is a structural feature of how Claude was trained, not a coincidence.

Spago Beverly Hills has 190 total normalized response appearances. ChatGPT accounts for 62; Claude for 61. One appearance separates the two leading platforms. That is the closest top-restaurant split in this dataset. What produces that near-perfect balance is not accident — it is the nature of a restaurant whose name has been canonical for over four decades. Wolfgang Puck's flagship opened in 1982. It has been in every dining guide, every travel feature, every "best of Los Angeles" list since. Every platform learned the same thing about Spago because the same thing has been written about Spago in every source that training data could absorb. The legacy brand achieves cross-platform consensus not by being remarkable but by being unignorable. Gemini trails at 44; Perplexity at 23.

Gemini leads or strongly features four restaurants in this dataset: 71Above, Hayato, Kato, and Somni. That list is not random. 71Above is a restaurant on the 71st floor of the Wilshire Grand Center — and Gemini accounts for 41 of its 60 total appearances. 68% of a restaurant's coverage concentrated on one platform signals a structural difference, not a random one. View-based dining content saturates travel guides and "dining with a view in Los Angeles" roundups. Gemini, drawing from Google-adjacent search sources, absorbs that content heavily. Hayato's omakase earns Gemini placement through dining guide lists. Kato's modern Chinese-American tasting menu appears in the search-accessible roundups Gemini reads. The Gemini map of Los Angeles fine dining is, in part, a map of what shows up in search results — not what shows up in long-form culinary criticism.

Mélisse has 135 total normalized response appearances across 600 responses. Perplexity leads at 40, narrowly ahead of Claude's 35. Gemini has 31; ChatGPT has 29. This is the only restaurant in the top seven where Perplexity holds the leading platform position — and the difference is instructive. Mélisse (in its current Citrin tasting-counter format) holds a Michelin 3-star rating. Its 2021 restructure generated active critical coverage and reservation discourse. That combination — credentialed, recently active in the press, with ongoing citation trails — is exactly the profile that Perplexity's live-web architecture rewards. Where the citation trail is active, Perplexity leads. Mélisse has one of the more active trails in the dataset.

• Mélisse is the only restaurant in the top seven where Perplexity outperforms both Claude and ChatGPT. Across the rest of the top tier, Claude or ChatGPT leads.
• The Michelin 3-star documentation creates a citation trail Perplexity follows regardless of culinary-writing depth. Awards matter to retrieval systems.
• Mélisse's post-2021 reformat press keeps the citation trail active — this is a restaurant that has been written about recently, not only historically.
• The pattern: Perplexity follows the live-web citation trail. Where that trail is active and award-backed, it leads. Mélisse is the clearest example in this dataset.

Kato and Nobu Malibu are tied at rank 10 in this dataset — both with 84 normalized response appearances. Their platform profiles are, structurally, mirror images. Kato: Gemini leads at 33, Perplexity follows at 24, Claude has 16, ChatGPT has 11. Nobu Malibu: Claude leads at 52, ChatGPT has 19, Gemini has 10, Perplexity has only 3. The same total. Completely different distributions. Kato is a modern Chinese-American tasting menu in West Adams with strong dining-guide placement — the kind of restaurant that appears in search-accessible roundups that Gemini and Perplexity read. Nobu Malibu is a beach-location restaurant whose culinary mythology, scenic setting, and celebrity association have been written about extensively in culinary and lifestyle press — the kind of coverage Claude absorbs. The ranking metric is the same. The machines that produced it are not.

The tie at rank 10 is one of the clearest illustrations in this dataset of what normalized response appearances actually measure: not the quality of a restaurant, but how many different machines, drawing from different knowledge architectures, surface it for the same queries. Kato and Nobu Malibu surface for different machines on different prompts and end up at the same number. The number looks identical. The story behind it is entirely different.

The restaurants AI misses in Los Angeles follow a pattern. Platform blind spots are not random — they are structural, shaped by each machine's knowledge architecture. Claude misses the view-based and search-saturated restaurants. Gemini misses the quietly critical, write-up-heavy restaurants that aren't prominently placed in location-based content. The restaurants that fall between platforms — recognized by one or two, ignored by the others — are the places where a diner using only one AI source would receive meaningfully incomplete recommendations.

• Orsa & Winston (45 total) — Claude: 1 appearance. Josef Centeno's Arts District Japanese-French tasting counter earns Gemini coverage (25) through omakase dining guide placement. For Claude, this restaurant is effectively invisible — one of the lowest Claude counts for a restaurant with serious culinary credentials in this dataset.
• 71Above (60 total) — Claude: 2 appearances. The 71st-floor restaurant is dominated by Gemini (41). The cuisine is secondary to the view in Gemini's coverage, and Claude doesn't surface view-based dining as a category. A restaurant that generates extensive search content and almost no culinary criticism.
• République (126 total) — Perplexity: 17 appearances. Walter Manzke's La Brea restaurant has 48 Claude appearances and 42 ChatGPT appearances, but only 17 in Perplexity. The culinary-writing footprint is significant; the live-web citation trail is not. Perplexity's thinner coverage of République is a reliable way to misunderstand it.
• The pattern: Restaurants with strong culinary-writing depth but thin search footprints appear clearly on Claude and are missed by Gemini and Perplexity. Restaurants with strong location and experience content appear on Gemini and are missed by Claude. The gap between what the machines know is not noise — it is the map of how each machine was trained.