ChatGPT is about to get Yelp's reviews. We measured what it cited before the deal: Yelp, zero times out of 511 answers

What was actually reported, and by whom

Quoting the Axios report rather than the write-ups of it: "Yelp is licensing reviews, photos and other business information to OpenAI, giving ChatGPT a new source of real-time local recommendation data, the company tells Axios exclusively." ChatGPT "will surface Yelp reviews, ratings, photos and other business details as part of responses to user queries." Yelp's branding and links will appear when OpenAI uses the content, "though OpenAI controls the design of that experience," per Stoppelman. Yelp's "Request a Quote" flow is to be added "soon," letting ChatGPT users contact local service providers inside the model. Terms were not disclosed, and the deal does not stop Yelp signing similar ones. Stoppelman's framing: "If you want to answer local queries, you really need Yelp."

A sourcing note we think belongs in the post, not the footnotes. This is single-sourced by design — an exclusive. We checked before writing: as of today, Yelp's own press-release index carries nothing about it (its only July 23 release announces the date of Q2 earnings), Yelp's newsroom has published nothing since June 9, and OpenAI has announced nothing. Everything else in circulation traces back to Axios. Figures being repeated downstream — a review count, a listing count, a share-price move — are not in the Axios piece and we could not verify them against a Yelp primary source, so we do not repeat them. Axios also discloses its own licensing agreement with OpenAI. No launch date has been stated by anyone. Treat this as reported-and-confirmed, not shipped.

The pre-deal baseline, from our own production corpus

Every audit we run asks the engines a business's own buyer questions — "who do you recommend for [category] in [city]?" — never the business's name. This cut is 173 completed audits, 1,024 answers (64 med spas, 42 personal-injury law firms, 40 dentists, 27 home-services businesses; 42 US metros; run 2026-06-12 to 2026-07-14), with synthetic smoke-test rows excluded by exact business name. For every ChatGPT answer we extracted every link and collapsed it to a registrable domain. The unit is answers-citing-a-domain, not links.

Zero is a strong claim, so here is why we trust it: those same 511 answers carried thousands of links across 1,092 different domains, including plenty of small one-off local blogs. The instrument plainly captures ChatGPT's links. Yelp is not missing from our measurement — it is missing from ChatGPT's answers. What does show up in its place is a stack of review aggregators and directories, several of which republish or re-rate the same underlying reputation data: Birdeye in a quarter of answers, the BBB in a sixth.

Why the zero exists — check it yourself in one command

This is not a ranking story. We fetched yelp.com/robots.txt today, 2026-07-25. Yelp gives OAI-SearchBot — the crawler behind ChatGPT's search results — `Disallow: /`, with narrow exceptions for `/article` and `/spk/`. `ChatGPT-User`, `PerplexityBot`, `Claude-SearchBot` and friends get the same treatment; `GPTBot`, `ClaudeBot`, `CCBot`, `Google-Extended` and `anthropic-ai` get a flat `Disallow: /`. Business pages — the `/biz/` URLs that hold the reviews — are closed to all of them. Googlebot, by contrast, has explicit `Allow:` rules. The file's own header states that automated access is "prohibited, except as expressly permitted by Yelp."

A licensing deal is exactly that express permission. So the change being reported is not that Yelp will start ranking better inside ChatGPT — it is that a corpus which was structurally unreachable becomes reachable, through a commercial pipe rather than a crawl. That distinction decides what anyone can do about it.

What this means if you sell or buy AI visibility

• There is no optimisation surface here, and you should not sell one. Licensed data arrives through a contract, not a crawler. No amount of schema, content or internal linking on a client's own site influences whether ChatGPT pulls their Yelp page. The actionable work is unglamorous data hygiene on the Yelp listing itself — correct hours, categories, address, photos, and a live review flow — because if that record becomes an input to ChatGPT, its errors become the answer's errors.
• Baseline now or lose the ability to attribute anything in Q3. The pre-deal state is measurable today and gone the moment this ships. If a local client's AI visibility moves in the next quarter and you have no dated before-line, you cannot distinguish your retainer from OpenAI switching on a data feed — and neither can your client, which is the more expensive half of that problem.
• Watch the cited-source mix, not a score. The measurable signature of this deal landing is compositional: yelp.com appearing in answers where it previously never did, and a fall in the aggregator pile currently filling that role. That is a real, checkable event. A single "visibility score" ticking up tells you nothing about what caused it.
• Yelp reputation already leaks into answers without a link. 11.0% of ChatGPT's answers discussed Yelp ratings and reviews sourced from third-party pages quoting them. Businesses treating Yelp as a dead channel because it sends no traffic have been wrong about that for a while; the engines were reading about it either way.

Limits

• This is a report about a deal, not an observed product change. We have measured no Yelp content in ChatGPT answers and are not claiming any. Our 0-of-511 is the state before, which is the only thing anyone can honestly measure right now.
• ChatGPT only. Our 513 Perplexity answers mention Yelp at a much higher rate — 180 of 513, 35.1% — but we capture Perplexity's text through the API rather than its citation panel, so it returned only a handful of link domains across the whole corpus. We can report Perplexity's Yelp mentions; we cannot report its Yelp citations, and we are not going to pretend otherwise.
• Convenience sample, single-sampled per question. 173 businesses we chose to audit across four verticals and 42 US metros, June 12 to July 14, 2026. A different vertical mix — restaurants above all, where Yelp is strongest — could look different. We audit no restaurants.
• We sell this measurement. AskedAbout sells AI-visibility audits, so check the method rather than the conclusion. The robots.txt claim takes one `curl`; the corpus claim you have to take from us, which is why the method is written out above.

You can run the free 60-second check on any local business and read the actual source list out of the answers — including whether anything of yours is in it today. Agencies who want a dated pre-deal line across a book of local clients can run five at once with the $249 Agency 5-pack — 25 buyer questions × 4 engines × 3 samples per business, white-labeled. The corpus-wide version of this cut is where AI looks before it recommends a local business.