AI Won’t Solve Yelp’s Problems

Must read

On January 30, Yelp announced that it too had succumbed to the allures of large language models, aka “AI.” In a press release, chief product officer Craig Saldanha says the review website is adding 20 new updates, including “AI-powered business summaries that make it even easier to find a business that fits your needs with a quick glance.” Those summaries basically take existing reviews and distill them into a few sentences. They won’t be scrummed from just any reviews, Yelp’s senior vice president of consumer product Akhil Kuduvalli Ramesh tells the Verge, but a “reliable number of recent reviews” recommended by software Yelp already uses.

Will this make Yelp more helpful? Maybe. Though the example Yelp chooses to highlight in the press release makes it seem like the summaries will be more redundant than anything. A summary for a place called “Darwin’s Diner” notes its cheeseburgers and affordability. Cheeseburgers? For a diner? Groundbreaking.

Yelp acknowledges that it’s been using AI “for years” in a variety of features to “help reduce consumers’ decision paralysis” when it comes to choosing restaurants and bars. Mainly, AI helps highlight commonly mentioned dishes in reviews. But the decision paralysis might be coming from inside the house. Using Yelp recently has felt like a minefield of often unrelated sponsored content (a recent search for “Chinese food” brought up Panera Bread), unhelpful community questions, and bad-faith commentary that makes it difficult to figure out if any given place is worth your time and money. At this point, 20 years in, is Yelp working?

A database of every business was always going to be unwieldy, but Yelp began with a decent idea: a place where everyone could air their opinions and experiences equally. This allowed locals to champion restaurants they loved that may have been overlooked by local media, or give a “real” assessment of places that had been met with perhaps unearned hype. In the early days, this created a powerful community of restaurant enthusiasts, some of which became part of the Yelp Elite Squad, verified users whose reviews are supposedly more trusted.

Of course, there were problems. Yelp Elite members pressured restaurants for free food or cash in exchange for good reviews, a trend that’s continued with Instagram and TikTok influencers. Then came the proliferation of “Yelp-bombing,” or reviews piled onto a restaurant many reviewers have never even frequented, based on media coverage, like when Demi Lovato’s Instagram complaint about frozen yogurt shop the Bigg Chill led to a flurry of both positive and negative reviews that had little to do with the quality of the froyo. And by this point, everyone has seen a baffling review of a beloved restaurant. Someone gave one of my favorite local restaurants one star because it didn’t have moscato on the wine list and then they had an allergic reaction to shellfish they ordered. Hope you’re okay, but that’s a you problem!

Summaries of these reviews generated by large learning models (LLMs) may help users know whether an Italian restaurant does indeed serve pasta. But the flawed assumption that Yelp makes, and sells, is that equal weight to all opinions will somehow give you the truth about a restaurant. The reasoning seems to go that one opinion is one opinion, but a thousand opinions is now data – which is of course science, fact, incontrovertible proof that whatever accumulated rating is accurate and deserved.

Yes, if you see a restaurant with mostly one-star reviews (that actually talk about the food), maybe go somewhere else. But the fundamental truth about restaurant reviews, whether they come from Yelp or TikTok or a James Beard Award-winning critic, is you’ll never have their meal. A good critic can invite you into their experiences with their writing, explaining how and why their meals succeeded or not in their estimation, drawing on experience and expertise. A group of 100 reviews from different diners can show you whether or not a critical mass of people liked a dish or not. But information is not the same as knowing, and you will not know whether you will enjoy yourself or not until you do it. At a certain point you just have to choose to go. And a two-sentence AI summary probably won’t help much.

More articles

Latest article