Reviews Regulate the Dining Room. Not the Kitchen.

Based on the research ofFarronato and Zervas, "Consumer Reviews and Regulation: Evidence from New York City Restaurants," Marketing Science, 2026

Chiara Farronato and Georgios Zervas paired Yelp review text with New York City restaurant inspection records and asked a question that sounds simple: does what people write about a meal actually track what an inspector finds. The honest answer is more useful than yes or no. Reviews are informative about the hygiene dimensions consumers directly experience, food temperature and pests above all, and far less informative about the dimensions that only a trained inspector can catch. Demand responds causally to the visible signal, and restaurants that are more exposed on Yelp turn out to violate less, but specifically on the dimensions where reviews carry information. Crowd oversight is real. It is also, provably, half of the job.

NYC's Letter Grades: The System Reviews Sit Beside

The paper's setting is not incidental. Since 2010, New York City has required every restaurant to post a letter grade tied to its most recent sanitary inspection score: 0 to 13 points earns an A, 14 to 27 a B, and 28 or more a C, displayed on a card at the door where anyone walking in can see it. The city's Department of Health and Mental Hygiene runs unannounced inspections of the roughly 29,000 restaurants under its jurisdiction at least once a year, and Farronato and Zervas draw on data the department shared directly with them. That official system is what makes the comparison possible in the first place: NYC's inspectors keep checking every dimension of the health code, visible or not, independent of whatever is happening on Yelp, so the researchers have a ground truth to hold reviews against.

That ground truth is granular in a way a letter grade is not. A single inspection score bundles violations with very different visibility profiles into one number: a diner will notice a plate that arrives cold, and will never see whether the walk-in cooler behind the kitchen door is holding at the right temperature between deliveries. The letter grade treats both as one input to a score. The paper's contribution is prying that bundle apart and asking which pieces of it show up in what customers actually write.

What Yelp Sees, and What It Doesn't

The finding cuts against the two intuitions people usually reach for. One says government inspection and crowd reviews are redundant, so a platform score should track an inspection score closely across the board. The other says reviews are just noise, a popularity contest dressed up as a quality signal, unreliable next to a trained inspector's checklist. Neither holds up. Reviews turn out to be well calibrated, but only on the dimensions an ordinary customer can personally detect with the senses they bring to a meal. Ask a review mentioning temperature or pests to predict an inspector's finding on those same categories, and it tracks. Ask it to say anything useful about a category that requires walking into the kitchen and checking a log, and it mostly has nothing to say.

The causal piece matters as much as the correlational one. Consumers are not just leaving cold-food complaints after a bad inspection becomes public; their demand moves in response to the hygiene signal embedded in reviews, and restaurants that are more exposed to that scrutiny on Yelp end up cleaner, specifically along the dimensions where reviews are informative. That is a real enforcement channel, running in parallel with the city's inspectors and independent of them. It is also a channel with a hard boundary, and the boundary sits exactly where a customer's senses stop being able to reach.

Yelp Built the Handshake, Not the Substitute

Yelp's own product history reads like an admission of this limit, even though the company built it years before this paper existed. In 2013, Yelp introduced LIVES, an open data standard for digitizing official restaurant hygiene scores and attaching them to business pages. A later study of Louisville, Kentucky's rollout, published in the Journal of Industrial Economics, found that displaying that inspection information on Yelp significantly reduced rates of severe food poisoning relative to the rest of the state. In 2022, Yelp expanded the program through a partnership with the Seattle-based firm Hazel Analytics, since acquired by Ecolab and rebranded Health Department Intelligence, which now powers hygiene data on nearly 700,000 Yelp pages, drawing on health departments across 48 states and, for the first time, Toronto and Vancouver, covering jurisdictions representing close to 70 percent of the U.S. population.

Notice what Yelp did not do. It did not decide that its own review text was informative enough to stand in for an inspection score. It went and grafted the government's actual number onto the page instead, sourced straight from the health department. Yelp's own account of the LIVES program traces the idea back further, to Ginger Zhe Jin and Phillip Leslie's 2003 study of Los Angeles County's 1998 mandate that restaurants post physical grade cards, a policy Yelp credits, citing that research, with roughly a 13 percent drop in hospitalizations for foodborne illness. The lineage is disclosure regulation first, crowd review second, official score reattached on top of both. A platform that had genuinely concluded reviews covered the whole hygiene picture would not have spent a decade building infrastructure to import the government's number back in.

A five-star Yelp page proves your dining room. It says nothing about your walk-in cooler's thermostat.

The False Comfort of a Clean Yelp Page

The risk this paper puts a number on is a specific kind of miscalibration: treating a strong Yelp presence as a general hygiene guarantee rather than a partial one. A restaurant can generate genuine, causally earned five-star momentum, hot food on time, no visible pests, and still be cutting corners entirely out of a customer's reach, on refrigeration discipline, cross-contamination protocol, whatever the inspector's clipboard covers that a dining room table never will. The two systems look substitutable from a distance. They are not. They are complementary, and lopsided, and the paper can only show that because New York never stopped running mandatory inspections independent of what Yelp was doing. Any city that leaned on a healthy volume of good reviews to justify a lighter inspection schedule would be quietly defunding the only channel watching the half that reviews cannot see.

What This Means for Anyone Designing Oversight

The pattern generalizes past restaurants to any setting where a platform's crowd signal sits next to a formal certification: a passenger's star rating next to a driver's background check, a guest's review next to a short-term rental's fire-code inspection, a parent's comment next to a childcare license. In every one of those pairs, the crowd signal is strong exactly where the customer has direct sensory or experiential access, and close to worthless on anything that requires specialized access, testing, or training the customer never gets. The job for anyone designing oversight is not to pick a side between reviews and inspectors. It is to map, dimension by dimension, which half of the checklist a customer could plausibly generate signal on, and to keep the formal channel's funding and coverage tied to exactly the dimensions the crowd cannot reach, because left unmanaged, a good enough average rating will eventually be trusted for all of it, at precisely the moment that trust is unearned.

Before you let a five-star page stand in for an inspector's clipboard, ask which line items on that clipboard a customer could never have written themselves, and staff that half as if the reviews never happened.