What the Machines Said

§ 01 QUINCE 627 mentions

01

Quince

627

total mentions

Michael Tusk has been doing the same thing quietly and brilliantly for twenty years. The machines noticed — all four of them, more than any other restaurant in San Francisco. What the consensus reveals is less about Quince and more about what AI considers a safe answer.

ChatGPT 156 mentions

Serves Quince as a search result: an address, a star rating, a hyperlink, and a '$$$$ French' category that has nothing to do with what Michael Tusk is actually cooking. One hundred and fifty-six mentions, each one formatted like a business listing.

Claude 170 mentions

One hundred and seventy bullet points of organized admiration. Claude processes Quince as a taxonomy — ⭐⭐⭐, Italian-California, seasonal sourcing, exceptional wine program. Everything correct, nothing unexpected.

Gemini 165 mentions

The most thorough brief in the room — cross-referenced, price-noted, occasion-tagged. Gemini recommends Quince with the confidence of something that has actually done its homework.

Perplexity 136 mentions

Cites sources rather than forming opinions. Quince appears because "sources consistently rank" it highly. That's a true statement and also a way of never having to say anything yourself.

§ 02 BENU 476 mentions

02

Benu

476

total mentions

Corey Lee spent decades finding a form for Korean diaspora cooking at the highest level of American fine dining. AI found the accolades. Whether it found the argument is another question.

ChatGPT 131 mentions

Returns Benu attached to a review count and a dollar sign tier. The first Korean-American to hold three Michelin stars arrives as a business listing with 4.6 stars and a price range.

Claude 120 mentions

Categorizes Benu as "Korean-influenced contemporary cuisine" and proceeds to enumerate. A chef who spent decades finding a form for diaspora cooking becomes a list of tasting menu attributes.

Gemini 118 mentions

Gives Benu the full write-up: the backstory, the culinary influence, the occasion recommendation. Of all the platforms, Gemini's response comes closest to describing what makes this restaurant distinctive.

Perplexity 107 mentions

Builds its recommendation from other sources' praise — accurate, well-referenced, somehow missing the point. Benu gets cited rather than described.

§ 03 ATELIER CRENN 458 mentions

03

Atelier Crenn

458

total mentions

Dominique Crenn has always had a point of view. Claude has the most to say about her — by a significant margin. Perhaps a chef who writes her menus as poems is the kind of thing a language model recognizes as its own territory.

ChatGPT 67 mentions

Only sixty-seven mentions — remarkable given Crenn's profile. ChatGPT's search grounding seems to favor restaurants with denser web footprints; three Michelin stars apparently make less noise in the index than expected.

Claude 142 mentions

One hundred and forty-two mentions, the highest Claude score of any restaurant in the dataset. Claude's depth on Crenn runs through her bio-dynamic farm, her poetic menus, her activism. It is, in its way, a fan letter.

Gemini 130 mentions

Recommends Crenn with the usual accuracy: Michelin stars, the farm, the tasting menu format. The recommendation is correct. The poetry doesn't quite survive the bullet point.

Perplexity 119 mentions

Documents Crenn thoroughly — the awards, the sustainability sourcing, the advocacy. Every claim cited. Crenn is not a chef who can be reduced to documentation, but Perplexity tries.

§ 04 GARY DANKO 456 mentions

04

Gary Danko

456

total mentions

Gary Danko has been the safe, impeccable choice for every corporate dinner and anniversary in San Francisco for twenty-five years. Perplexity leads here, which makes sense — Danko is the most citeable restaurant in the city.

ChatGPT 68 mentions

Sixty-eight mentions — fewer than the other platforms, and lower than Gary Danko's standing in the dining scene might suggest. ChatGPT's search-grounded responses produced fewer mentions of this restaurant than either Gemini or Perplexity.

Claude 98 mentions

Ninety-eight mentions of reliable competence. Claude recommends Gary Danko as you'd expect: formal, occasion-appropriate, technically impeccable. The same way the restaurant has operated every night for twenty-five years.

Gemini 133 mentions

One hundred and thirty-three mentions — Gemini clearly indexes special-occasion dining by institutional reputation, and Danko's reputation is bedrock.

Perplexity 157 mentions

One hundred and fifty-seven mentions — the highest score of any platform for any restaurant except Quince. Perplexity loves an institution: documented, sourced, consistent, never surprising.

§ 05 SAISON 397 mentions

05

Saison

397

total mentions

One of the most expensive meals in America, built around a single wood fire. ChatGPT and Gemini know it. Claude is still deciding — sixty-nine mentions is the softest endorsement in the top five.

ChatGPT 120 mentions

One hundred and twenty mentions for a $500 tasting menu. ChatGPT recommends Saison with the same even tone as any other result, which says something interesting about how it processes price.

Claude 69 mentions

Only sixty-nine mentions. Claude seems uncertain about Saison — perhaps the wood-fire ethos and the price point don't fit neatly into a category. There is no appropriate bullet for "one of the hardest restaurants to fully explain in America."

Gemini 125 mentions

Handles Saison correctly: the wood fire, the forager sourcing, the $500 floor. The description is accurate. The experience is not something a list describes, and Gemini doesn't pretend otherwise.

Perplexity 83 mentions

Eighty-three mentions — a lower citation footprint than the top-tier restaurants. Saison has been famously press-averse, and Perplexity's model feels the absence.

§ 06 LAZY BEAR 326 mentions

06

Lazy Bear

326

total mentions

Lazy Bear built its reputation through supper-club word of mouth and a ticketing model no one had seen before. ChatGPT's search grounding apparently operates on the same old-media channels Lazy Bear specifically avoided.

ChatGPT 32 mentions

Thirty-two mentions. ChatGPT missed the memo — or missed the method. Lazy Bear's communal dining, ticket-in-advance format, and chef-as-host energy exist largely outside the search layer.

Claude 70 mentions

Seventy mentions — fair treatment but less animated than the traditional Michelin establishments. The communal dining concept and the supper-club energy don't bullet-point naturally.

Gemini 120 mentions

One hundred and twenty mentions — the platform leader. Gemini describes the ticket-in-advance format, the communal tables, the menu-as-performance. Thorough and correct.

Perplexity 104 mentions

One hundred and four mentions. Sources its Lazy Bear recommendation well — the Eater write-ups, the Michelin stars, the awards — and presents it as a well-documented phenomenon.

§ 07 ACQUERELLO 284 mentions

07

Acquerello

284

total mentions

A Nob Hill institution that has operated with near-total exterior silence for thirty-five years. AI finally caught up — unevenly. Perplexity's low score (38) reflects a restaurant that doesn't generate the citation density the model needs.

ChatGPT 88 mentions

Eighty-eight mentions for a restaurant that has generated almost no noise for three decades. ChatGPT recommends it as "Italian fine dining," which is like recommending a Rothko as "rectangular canvas, oil."

Claude 60 mentions

Sixty mentions — competent and correct. The responses cover the format and credentials; the thirty-five years of risotto execution are not something any platform can convey from text.

Gemini 98 mentions

Ninety-eight mentions, the platform leader. Gemini clearly values institutional Italian fine dining, and Acquerello has thirty-five years of institutional weight to offer.

Perplexity 38 mentions

Thirty-eight mentions — the lowest of any platform. Acquerello is not a restaurant that generates the kind of online citation density Perplexity feeds on. The food remains indifferent to this.

§ 08 KOKKARI 253 mentions

08

Kokkari

253

total mentions

The best Greek restaurant in San Francisco has been on Jackson Street for twenty-five years. Claude has almost nothing to say about it. This is not a restaurant that lives in the places AI tends to look.

ChatGPT 25 mentions

Twenty-five mentions. Kokkari is beloved by every chef in the city and nearly invisible to the algorithm that talks to the most people. The gap between reputation and search-layer presence has never been more legible.

Claude 10 mentions

Ten mentions. Claude's near-silence on Kokkari is the most revealing score on this list. A restaurant that has defined Greek-Californian cooking for a quarter century — and one of the most widely used platforms surfaces it only ten times.

Gemini 116 mentions

One hundred and sixteen mentions — the platform leader and the most accurate read of Kokkari's actual local standing. Gemini's broader sourcing found what the others missed.

Perplexity 102 mentions

One hundred and two mentions — second-highest. Perplexity sources the Kokkari story well from local food press and recommends it with confidence. The citation record, it turns out, is there if you know where to look.

§ 09 CALIFORNIOS 227 mentions

09

Californios

227

total mentions

Val Cantu's Mexican fine dining tasting menu was ahead of its time when it opened — and the platform data reflects continued unevenness in how it's classified. Gemini leads by a wide margin at 90 mentions. The source of this platform gap is not established by this dataset; one hypothesis is that Gemini draws from broader food press coverage, though this has not been directly tested.

ChatGPT 43 mentions

Forty-three mentions — fewer than Gemini, with responses that reflect less settled categorization. Mexican fine dining tasting menu is a category the industry took years to accept, and this platform's handling of Californios suggests the categorization is not yet consistent.

Claude 35 mentions

Thirty-five mentions. Claude gives Californios appropriate handling but can't quite locate it in a familiar structure. The restaurant's essential strangeness — the form it invented — doesn't compress into a list.

Gemini 90 mentions

Ninety mentions — the leader by a significant margin. Gemini's broader sourcing picks up more of the food press coverage Californios has earned for defining a new category of American fine dining.

Perplexity 59 mentions

Fifty-nine mentions. Sources the story correctly — the James Beard nominations, the tasting menu, the Mission address. The citations are there. The argument for why this restaurant matters is harder to source.

§ 10 RICH TABLE 213 mentions

10

Rich Table

213

total mentions

Evan and Sarah Rich built the definitive Hayes Valley neighborhood restaurant. AI agrees — it leads every Neighborhood cluster query in the city. Perplexity's lead suggests the restaurant's role as a neighborhood anchor generates a particular kind of citeable documentation.

ChatGPT 18 mentions

Eighteen mentions. Rich Table is a neighborhood restaurant that happens to be excellent, and ChatGPT seems to evaluate it only as the former. Neighborhood dining doesn't surface in the same queries as Michelin temples.

Claude 31 mentions

Thirty-one mentions. Claude identifies Rich Table correctly as a Hayes Valley institution, but the word count suggests it has less to say about creative Californian cooking than about three-star tasting menus.

Gemini 64 mentions

Sixty-four mentions. Gemini handles Rich Table well enough — the sardine chips, the seasonal sourcing — without quite conveying why chefs eat here on their nights off.

Perplexity 100 mentions

One hundred mentions — the platform leader by a wide margin. Rich Table leads local neighborhood dining queries, a category with its own documentable citation density distinct from Michelin press.

§ 11 BIRDSONG 193 mentions

11

Birdsong

193

total mentions

Christopher Bleidorn built something genuinely original in SoMa. Perplexity found it first. Search-grounded tools favor restaurants with dense press histories; Birdsong is too recent and too good to be famous yet.

ChatGPT 24 mentions

Twenty-four mentions. A restaurant earning serious critical attention barely registers. ChatGPT's search layer favors restaurants with long press histories, and Birdsong is still building its record.

Claude 33 mentions

Thirty-three mentions. Claude finds Birdsong and describes it accurately — the native ingredients, the forage-forward sourcing — but the restaurant's specific identity remains somewhat compressed.

Gemini 58 mentions

Fifty-eight mentions. The most useful platform brief here — cross-references the awards and the sourcing philosophy. Gemini is better at finding emerging restaurants than its competitors.

Perplexity 78 mentions

Seventy-eight mentions — the platform leader. Perplexity's citation model picks up the awards coverage and the early-adopter food press writing that Birdsong has started generating.

§ 12 STATE BIRD PROVISIONS 173 mentions

12

State Bird Provisions

173

total mentions

Stuart Brioza turned a dim sum cart into a James Beard Award and a new format for California dining. Twenty-five ChatGPT mentions for a restaurant that changed how San Francisco thinks about a meal.

ChatGPT 25 mentions

Twenty-five mentions for a restaurant that invented its own format. State Bird's dim sum service with California ingredient obsession doesn't map cleanly to search vocabulary, and the mention count shows it.

Claude 47 mentions

Forty-seven mentions. Claude identifies the "progressive dim sum" concept and the James Beard Awards. It gets the philosophy. Whether it gets why the restaurant matters culturally is a different question.

Gemini 54 mentions

Fifty-four mentions. Gives State Bird a solid brief — the cart-service format, the local sourcing, the occasion recommendation. Correct and present, though the essential strangeness of the experience is missing.

Perplexity 47 mentions

Forty-seven mentions. Sourced and accurate, but the restaurant's essential strangeness — the way a meal there surprises you — doesn't survive the citation model.

§ 13 WAYFARE TAVERN 156 mentions

13

Wayfare Tavern

156

total mentions

Tyler Florence's Financial District institution owns the corporate dinner conversation in this city. Six ChatGPT mentions. The deepest mystery in this audit — and possibly evidence that ChatGPT simply doesn't eat lunch in the Financial District.

ChatGPT 6 mentions

Six mentions. A restaurant that has defined SF power dining for a decade — six. There is no satisfying explanation for this score. ChatGPT's search grounding simply did not surface it.

Claude 42 mentions

Forty-two mentions. Claude recommends Wayfare Tavern for private dining and corporate occasions, which is precisely what it was built for. The recommendation is accurate. The food is better than the recommendation implies.

Gemini 46 mentions

Forty-six mentions. Handles Wayfare competently — the roast chicken, the private rooms, the American brasserie register. A reliable recommendation for a reliable restaurant.

Perplexity 62 mentions

Sixty-two mentions — the platform leader. The occasion-dining citations and the press history give Perplexity enough to work with. It works with them.

§ 14 THE PROGRESS 143 mentions

14

The Progress

143

total mentions

State Bird's quieter sibling. More ambitious, less discovered — by critics and AI alike. Gemini leads, which suggests it handles The Progress as a distinct restaurant identity rather than a footnote.

ChatGPT 10 mentions

Ten mentions. ChatGPT appears to have filed The Progress under "still State Bird." The menu is more ambitious than that filing suggests.

Claude 29 mentions

Twenty-nine mentions. Claude identifies The Progress correctly — the dinner-party format, the shared plates, the Fillmore location — and gives it more space than ChatGPT allowed. Still less than it deserves.

Gemini 55 mentions

Fifty-five mentions — the platform leader. Gemini handles The Progress as its own identity, not as State Bird's addendum. The most accurate brief here.

Perplexity 49 mentions

Forty-nine mentions. Good citation depth — the Brioza and Krasinski partnership, the format, the accolades. The Progress earns its Perplexity mentions through documented achievement.

§ 15 NOPA 139 mentions

15

Nopa

139

total mentions

A restaurant that has fed more off-duty chefs than anywhere in the city. Claude leads here, which is interesting — perhaps Nopa's reputation in food writing translates directly into training data density.

ChatGPT 10 mentions

Ten mentions. Nopa's reputation lives in the industry and in the neighborhood, not in search results. ChatGPT found very little of it.

Claude 49 mentions

Forty-nine mentions — the platform leader. Claude produces the most detailed Nopa descriptions of any platform; the source of that depth is not established from the outside, though the restaurant's longstanding presence in SF food writing is one plausible contributing factor.

Gemini 37 mentions

Thirty-seven mentions. Gives Nopa appropriate coverage — the late-night menu, the neighborhood anchor, the whole rotisserie chicken — but doesn't capture why it's indispensable.

Perplexity 43 mentions

Forty-three mentions. Sources the Nopa story accurately: the NoPa anchor, the farm-to-table ethos, the late hours. The citation record is there even if the soul of the restaurant resists documentation.

§ 16 BIX 127 mentions

16

Bix

127

total mentions

A supper club that feels like 1940s Shanghai filtered through North Beach. Claude leads by an enormous margin — seventy-two mentions against ChatGPT's two. A restaurant that self-identifies as literary, and the most literary AI noticed.

ChatGPT 2 mentions

Two mentions. A supper club that exists outside of time is apparently not the kind of answer ChatGPT's search grounding produces. Two is almost certainly an accident.

Claude 72 mentions

Seventy-two mentions — the platform leader by a margin that has no parallel in this dataset. Perhaps a restaurant that self-identifies as literary resonates with a model trained on literature. Perhaps Claude simply likes jazz.

Gemini 34 mentions

Thirty-four mentions. Describes Bix accurately enough — the cocktail culture, the live jazz, the special-occasion mood. What it cannot convey is the feeling of entering a room that exists outside of time.

Perplexity 19 mentions

Nineteen mentions. Bix doesn't generate the citation density Perplexity needs, and the restaurant appears to be entirely fine with that.

§ 17 KILN 116 mentions

17

Kiln

116

total mentions

Ron Siegel's wood-fire temple in the Presidio is earning serious attention — everywhere except Claude, which has given it exactly zero mentions. The most jarring absence in the dataset.

ChatGPT 54 mentions

Fifty-four mentions — the platform leader. ChatGPT leads on Kiln; the dataset establishes this platform gap but does not establish its cause. One hypothesis is that search-grounded responses drew from press coverage of Kiln's opening, though this has not been directly tested.

Claude 0 mentions

Zero mentions. For a restaurant earning the kind of critical attention that should generate AI coverage, Claude's complete absence is the most striking single platform gap in this dataset.

Gemini 38 mentions

Thirty-eight mentions. Finds Kiln with good sourcing — the wood fire, the location, the tasting menu format. A clean brief for a restaurant still building its name.

Perplexity 24 mentions

Twenty-four mentions. A smaller citation footprint than the critical attention suggests Kiln deserves. The press record is still being written.

§ 18 SONS & DAUGHTERS 113 mentions

18

Sons & Daughters

113

total mentions

Quietly excellent for years in a tiny Tendernob dining room. Nearly invisible on Claude despite holding its own everywhere else. The three-platform parity (37, 37, 36) is its own kind of editorial statement.

ChatGPT 37 mentions

Thirty-seven mentions. Sons & Daughters earns its ChatGPT presence through a consistent press record — search grounding rewards longevity and documentation, and the restaurant has both.

Claude 3 mentions

Three mentions. Nearly invisible — strange for a restaurant of this quality. Claude's sparse coverage suggests the restaurant's relatively quiet public profile hasn't translated into training data density.

Gemini 37 mentions

Thirty-seven mentions. Handles Sons & Daughters with appropriate depth — the intimate space, the progressive California menu, the tasting format. Accurate.

Perplexity 36 mentions

Thirty-six mentions. A remarkably balanced score across three platforms. Perplexity's citation count reflects a substantial and steady press record.

§ 19 FOREIGN CINEMA 106 mentions

19

Foreign Cinema

106

total mentions

Irene Buenviaje's Mission institution has been projecting films and serving oysters since 1999. The most balanced visibility profile on this list — broadly findable, consistently itself. ChatGPT and Claude give it the exact same score, which happens almost nowhere else.

ChatGPT 27 mentions

Twenty-seven mentions. ChatGPT recommends Foreign Cinema with appropriate awareness of both its functions — the food and the films — which is more than most platforms manage.

Claude 27 mentions

Twenty-seven mentions — identical to ChatGPT, which happens almost nowhere else in this dataset. Foreign Cinema apparently generates equal responses across model architectures. The restaurant has been doing the same thing for twenty-five years; the algorithms agree.

Gemini 33 mentions

Thirty-three mentions — the platform leader, by a small margin. Gemini recommends Foreign Cinema for brunch and dinner with equal confidence, correctly identifying the film projections as part of the dining experience.

Perplexity 19 mentions

Nineteen mentions — the lowest of any platform. The dual food-and-cinema identity may make the citation model uncertain which category to file it under. Foreign Cinema, as usual, refuses to be filed.

§ 20 EPIC STEAK 105 mentions

20

Epic Steak

105

total mentions

A steakhouse on the Embarcadero where the Bay Bridge view is half the restaurant. Gemini leads here — nearly half the total mentions — which tells you something about a platform that recommends by occasion and setting as readily as it does by food.

ChatGPT 20 mentions

Twenty mentions. ChatGPT recommends Epic Steak with the usual transactional directness — the waterfront address, the prime beef, the occasion framing. Respectable for a steakhouse competing in the same queries as the Michelin tier.

Claude 22 mentions

Twenty-two mentions — the closest to ChatGPT's score of any restaurant on this list. Claude gives Epic Steak the classic steakhouse brief: the Embarcadero setting, the special-occasion positioning, the views. Correct and unremarkable in equal measure.

Gemini 44 mentions

Forty-four mentions — the platform leader with nearly half the total. Gemini surfaces Epic Steak more than any other platform, which suggests it indexes destination dining and view restaurants with particular enthusiasm. When AI recommends Epic Steak, it is mostly recommending a view.

Perplexity 19 mentions

Nineteen mentions — the lowest of any platform. Epic Steak generates less of the food-critical citation density that Perplexity's model feeds on. It's a steakhouse with a view, not a tasting menu with a press strategy, and the score reflects that distinction precisely.