ChatGPT Reads Your Yelp Page Now. Yelp Chooses What It Reads.

On July 23, 2026, Axios reported that Yelp had licensed its review corpus to OpenAI. The agreement is non-exclusive, the financial terms were not disclosed, and it does something quite specific: when someone asks ChatGPT where to get dinner or which plumber to call, the answer can now arrive with Yelp ratings, reviews, photos and a link back to the listing.

Three weeks later, on August 10, Yelp went further and put reservations and waitlist joins inside ChatGPT itself. A customer can go from a vague question to a table without ever opening a browser tab.

Yelp's chief executive Jeremy Stoppelman put the case to Axios in one line: "If you want to answer local queries, you really need Yelp." He is not a neutral party. He is also, on this narrow point, describing something real.

Why OpenAI paid for this

Local questions have been a soft spot for ChatGPT since it launched. Search Engine Land put it plainly in its coverage of the deal: local prompts have been a weakness. The reason is structural rather than technical. An assistant can only answer confidently from sources it is allowed to read, and a licensed corpus that combines business details, photos, hours and a decade of written reviews is a rare thing to be able to buy.

Yelp already licenses that corpus to Apple Maps and to Yahoo, and the OpenAI agreement is non-exclusive, so it can sell it again. The one buyer not on the list is Google, which Yelp is currently suing over local search. Whatever else this deal is, it also lets Yelp route around Google by supplying Google's competitor.

None of which would matter to a restaurant owner, except for what it changes about who is reading their reviews, and how many of those reviews anyone gets to see.

The page it reads is not the page you think you have

Yelp does not publish every review it receives. It runs automated software that decides which ones to recommend. The recommended set is what appears on your page by default and what feeds your star rating. Everything else is pushed to a separate link at the bottom of the page, in smaller type, excluded from the rating, and read by almost nobody.

That system has existed for over a decade and has always been contested. What changed in July is the audience. It used to decide what a customer saw when they clicked through to your listing. It now shapes what a machine reads when a customer asks a question and never clicks anything.

Eleven out of ninety eight. That business has a Yelp page showing eleven reviews, a rating computed from eleven reviews, and now, potentially, an appearance in an AI answer built on the same eleven. Whether the other eighty seven were spam, solicited, written by people with thin Yelp histories, or simply unlucky, the owner has no way to audit it and no route of appeal.

This is the thing worth sitting with. The assistant is not forming an opinion about you. It is repeating one that Yelp already filtered.

The question nobody has answered

Here is what has not been made public: whether the corpus Yelp licensed to OpenAI contains only the recommended reviews, or all of them. Yelp has not said, and neither has OpenAI. Both readings are uncomfortable, in different directions.

If ChatGPT only sees the recommended set, then your AI reputation is a tidier version of your real one, which sounds like good news until you are the customer whose entirely genuine complaint was filtered out and is now invisible to the machine advising the next person. If it sees everything, then reviews Yelp itself declined to stand behind are feeding a paragraph a customer will read as a recommendation.

For an owner the operative fact is the same under either reading. A body of text you did not write, cannot fully see and have no ability to correct is now an input to how you get described. That is not new, exactly. Reviews have always worked that way. What is new is that the description now reaches the customer pre-chewed, in three sentences, with no ten blue links underneath to argue with it.

The trust gap is the part you can work on

The rest of the research is more encouraging than the headline suggests, and it points somewhere concrete.

Morning Consult found that 65% of Americans have used AI search, but only 15% trust it a lot, and 72% think AI platforms should always identify their sources. BrightLocal, surveying 1,002 US adults in March 2026, found that 45% now use AI to find a local business against 6% a year earlier, and that 88% of those people verify what they are told before committing. Just over half specifically go and check whether the reviews look legitimate.

So the assistant does not close the sale. It builds a shortlist of three, and then the customer does what they have always done: they go and read your reviews with a slightly sceptical eye. Being named gets you into the consideration set. What happens next is governed by the same unglamorous things it always was.

Which is where most owners are leaving money on the table, and it has nothing to do with AI. BrightLocal's February 2026 survey found that 89% of consumers expect owners to respond to reviews, that the share expecting a same-day answer tripled in twelve months, and that 42% are unlikely to use a business that ignores its reviews altogether. An AI recommendation delivered to someone who then lands on a page of unanswered one star reviews has done you no good at all.

The assistant is not forming an opinion about you. It is repeating one that Yelp already filtered.

What to do this week

Four things, none of which require buying anything. The first two take about ten minutes between them.

1. 01

   Read your own Yelp page as a stranger

   Open it in a private window and scroll to the very bottom. There is a link to reviews that are not currently recommended, with a count next to it. That number is your filtered share, and most owners have never once looked at it. Compare it to the review count you thought you had.

2. 02

   Claim the listing if you have not

   Plenty of small businesses have a live, populated Yelp page they never claimed, built from public data years ago. An unclaimed page can carry wrong hours and a dead phone number straight into an AI answer, and you cannot reply to anything on it.

3. 03

   Ask the question your customer would ask

   Open ChatGPT, type the query someone in your city would actually type, and read who gets named. Our free visibility check runs it repeatedly instead of once, because a single answer moves between runs. The list of names that comes back is your real competitive set, and it is usually not the one you assumed.

4. 04

   Answer the reviews that do show

   Response rate is the one input here you fully control, on every platform at once. It is also what the 88% who verify an AI recommendation are looking at when they arrive. Start with anything negative and unanswered from the last ninety days.

The uncomfortable part

A lot of American small businesses made a deliberate decision, years ago, to stop putting effort into Yelp and put all of it into Google. Measured on traffic, that decision was correct, and the raw numbers still favour Google today.

What has quietly changed is that the neglected page is no longer just a low traffic listing. It is a licensed data source feeding ChatGPT, the assistant named most often by the 45% of consumers who now use AI to find a local business, and it is doing that whether or not anyone at the business has logged in since 2019. The same logic applies to not having a website: the asset you skipped was fine to skip while humans were the only audience.

You do not need to fall in love with Yelp. You need to know what it currently says about you, across all four platforms rather than the one you check, which is the boring version of watching more than one dashboard. The machine is reading. It would be good to have read the same thing first.

Questions people actually ask