ChatGPT and Yelp: what the deal actually changes for local businesses
Yelp licensed its reviews, ratings, and photos to OpenAI on 23 July 2026. Every write-up so far has covered the announcement. Nobody has checked the one thing that decides what you can do about it: whether an AI assistant can reach a Yelp business page on its own. It cannot, and that is what makes this deal different from a normal source.
Bottom line
Yelp's robots.txt blocks every named AI crawler and agent from every business page, and was still doing so on 27 July 2026, four days after the OpenAI deal was announced. So 0% of Yelp's business pages can be reached by retrieval. The licensed feed is the only route into ChatGPT, which means you cannot optimise your way in, you can only make sure the record Yelp holds about you is complete and correct. Two more numbers set the size of the prize. Only 76% of the widely quoted 330.2 million reviews are recommended and count toward a star rating, and 68% of Yelp's 2025 advertising revenue came from services categories rather than restaurants, retail and hospitality. This is a large upgrade for a plumber and a small one for a hotel.
Confidence: the crawler figure is a direct read of yelp.com/robots.txt retrieved on 27 July 2026, and robots directives describe permission, not enforcement. The review and revenue figures are from Yelp's FY2025 Form 10-K, audited and dated 31 December 2025. Neither company has disclosed which records OpenAI receives, so the corpus figures describe Yelp, not the contract.
What was announced
On 23 July 2026, Axios reported that Yelp had licensed its reviews, ratings, photos, and business information to OpenAI for use in ChatGPT. Coverage puts the corpus at 330 million cumulative reviews and more than 8 million business listings. Search Engine Land reported that answers using the content will carry Yelp branding and links, with OpenAI deciding how that is presented. Yelp also plans to bring Request a Quote into ChatGPT, so a user can message a local business without leaving the chat.
Financial terms were not disclosed. The agreement is non-exclusive, so Yelp can sell the same content to other AI companies. Yelp shares rose about 8% on the day. Yelp's chief executive Jeremy Stoppelman told Axios that if you want to answer local queries, you really need Yelp.
All of that is the announcement. The useful question is what it changes for a business that wants to be the one ChatGPT names, and answering that requires looking at how Yelp content can and cannot travel.
Yelp still blocks every AI crawler from every business page
We fetched yelp.com/robots.txt on 27 July 2026, four days after the announcement. It runs to 336 lines and it is an allowlist: the final rule is a blanket disallow for any agent not named earlier. Googlebot, bingbot and a short list of social preview bots are let in. Every AI crawler and every AI agent is not.
| Agent group | Allowed | Blocked |
|---|---|---|
| OAI-SearchBot | /spk/, /article | Everything else, including all /biz/ pages |
| ChatGPT-User, Claude-User, Claude-SearchBot, PerplexityBot, Perplexity-User | /article | Everything else, including all /biz/ pages |
| GPTBot, ClaudeBot, CCBot, Google-Extended, anthropic-ai, Claude-Web, Meta-ExternalAgent | Nothing | The whole site |
| Everything not named (User-agent: *) | Nothing | The whole site |
Read from yelp.com/robots.txt on 27 July 2026. The narrow allows cover Yelp's own editorial articles at /article and its /spk/ paths, not business listings.
The relevant lines, verbatim
User-Agent: OAI-SearchBot
Allow: /spk/
Allow: /article$
Allow: /article/
Disallow: /
User-Agent: GPTBot
User-Agent: ClaudeBot
User-Agent: CCBot
User-Agent: Google-Extended
User-Agent: anthropic-ai
User-Agent: Claude-Web
User-Agent: Meta-ExternalAgent
User-Agent: Meta-ExternalFetcher
Disallow: /
User-Agent: *
Disallow: /This is the fact that changes how you should read the deal. A crawlable source is one you can influence: you publish something, an assistant fetches it, and what you wrote can end up in an answer. Yelp is not that. There is no path from your Yelp business page to an AI assistant except the one Yelp sells. Whatever reaches ChatGPT reaches it because Yelp put it in the pipe, in the shape Yelp chose, in the order Yelp's own software decided.
So the usual advice for getting cited by AI assistants does not transfer here. You cannot make your Yelp page more crawlable, more quotable, or better structured for retrieval, because nothing is retrieving it. The only surface you control is the record itself.
When a source is crawlable you can influence what an assistant reads. When it is licensed and closed, the licensor decides what the assistant receives. Yelp just moved local discovery from the first category to the second.
330 million reviews is not the number that carries a verdict
Every report on the deal repeats 330 million. That is the count of reviews ever submitted. Yelp's own annual report breaks it down differently, and the breakdown matters, because Yelp does not treat all of those reviews as reliable and does not let all of them affect a star rating.
| Measure, as of 31 December 2025 | Reviews | What it means |
|---|---|---|
| Reviews submitted to the platform | 330.2m | Up 7% from 308.1m at the end of 2024. This is the number in the press coverage. |
| Available on business pages | 300.3m | The other 29.9m were removed, either by Yelp for terms of service violations or by the users who wrote them. |
| Shown but not recommended | 48.1m | Yelp's recommendation software moves these to secondary pages, and they do not factor into the business's star rating. |
| Recommended, and behind the star rating | 252.2m | 76% of the headline number. This is the corpus that carries Yelp's actual verdict on a business. |
Yelp Inc. FY2025 Form 10-K, filed 27 February 2026. The 252.2m figure is the arithmetic: 300.3m available on business pages less 48.1m not recommended.
Yelp describes its recommendation software as automated trust and safety systems that analyse the relevance, reliability and utility of each review, using hundreds of signals and billions of data points. Reviews it does not recommend go to secondary pages and do not factor into the star rating, and no employee, business owner, or reviewer can override it. That filter now sits upstream of an AI answer.
Yelp has not said which slice of the corpus OpenAI receives. But the practical reading is the same either way. If ChatGPT reflects Yelp's ratings, it inherits Yelp's filtering. Review campaigns that generate a burst of submissions from new accounts are what the filter is built to catch, and they add to the 48.1 million rather than to your rating. Steady, unprompted reviews from established Yelp users are what survive it.
Yelp is a services company now, and that decides who this helps
Yelp reports its advertising revenue in two groupings, and the split has been moving in one direction for years. In 2025 services produced 68% of advertising revenue and grew, while restaurants, retail and other shrank.
| Grouping | Categories Yelp puts in it | 2025 ad revenue | Year on year |
|---|---|---|---|
| Services | Home, local, auto, professional, pets, events, real estate and financial services | $947.6m | +8% |
| Restaurants, retail and other | Restaurants, shopping, beauty and fitness, health, and other | $443.7m | -6% |
Yelp Inc. FY2025 Form 10-K. Total advertising revenue was $1.39bn, up 3%. Paying advertising locations averaged 510,000 across the year, down 3%.
Read the category lists carefully, because there is something missing. Services is home, local, auto, professional, pets, events, real estate and financial services. The other grouping is restaurants, shopping, beauty and fitness, health, and other. Lodging is not named anywhere. Hotels, short-term rentals, and coworking spaces sit in the residual of a grouping that is shrinking, inside the half of the business Yelp is not leaning on.
That is the honest frame for this deal. It is a meaningful upgrade to ChatGPT for home services, restaurants, salons, clinics, and auto repair, where Yelp's coverage is deep and commercially maintained. For a hotel or a coworking operator it is a marginal improvement to a source that was never strong for the category.
What to do if you run a services or hospitality business Yelp covers well
Restaurants, home services, salons, clinics, auto repair, retail with a storefront.
Treat the Yelp profile as a data feed, not a listing
Every field you leave blank is a field that cannot reach ChatGPT. Claim the page, fill in hours, service area, categories, licences, price range, payment methods, accessibility, and photos. OpenAI receives what Yelp holds, so the profile is now upstream of an AI answer.
Get Request a Quote working before it lands in ChatGPT
Yelp plans to bring Request a Quote into the ChatGPT answer, so a user can message you without leaving the chat. If the feature is off, or nobody answers it, you are advertising a dead end in a new channel. Turn it on and set a response-time target.
Work on recommended reviews, not review count
48.1 million reviews sit on Yelp's secondary pages and do not touch the star rating. Soliciting a burst of reviews from new accounts is the fastest way to add to that pile. Steady, unprompted reviews from established Yelp users are what survive the filter.
Stop counting on crawlability for Yelp content
Your own site still needs to be crawlable. Your Yelp page never will be. Anything you want an assistant to read about you from Yelp has to be in the structured fields Yelp licenses, not in a long profile description you hope a bot will find.
What to do if you run a hotel, a rental, or a coworking space
The short answer is: do not reprioritise around this. Yelp is not where your category is strong, and the deal does not create a route you can work on. Spend the effort on the sources that serve your guests and are still crawlable, and keep your Yelp record accurate so it cannot hurt you.
Check whether Yelp is even your category
Yelp reports its business in two groupings: services, and restaurants, retail and other. Lodging is not named in either. Hotels, vacation rentals, and coworking sit inside the residual, and they are not where Yelp's commercial attention or its freshest data goes. Search your own city on Yelp and count how many comparable properties have more than a handful of recommended reviews. That is your realistic ceiling.
Put the effort into the sources that do serve your category
For a hotel, the corroborating set is travel media, review platforms, maps, tourism boards, and your own crawlable neighbourhood and amenity pages. For a coworking space it is local business media, maps, and directories. Those are still fetchable by AI crawlers, which is exactly what Yelp is not.
Fix the fields the feed can carry anyway
A thin Yelp page is still a page an assistant can now be handed. Wrong hours, a dead phone number, or a closed-location duplicate are worse in an AI answer than on a listing nobody reads, because there is no results page next to it to correct the error.
Measure the answer, not the listing
The only way to know whether this deal moved anything for your property is to ask the assistants the questions your guests ask, on a schedule, and record what they name and cite. A listing audit cannot tell you that.
We have written per-category playbooks for the source sets that do work in these verticals: AI search for hotels, AI search for vacation rentals, and AI search for coworking spaces.
The wider pattern: closed sources with paid pipes
Yelp is not unusual in blocking AI crawlers. It is unusual in blocking them and then selling the same content through a contract. That combination is becoming the standard shape for large user-generated corpora, and it splits the web an assistant sees into two halves: an open half you can influence by publishing, and a licensed half where a company decides what the model gets.
For anyone doing AI search work, the practical consequence is that source audits have to record which half a source is in. A source you can be cited from and a source you can only be listed in call for completely different work. It also means you cannot infer what ChatGPT knows from what is crawlable, which is why the only reliable measurement is tracking what the assistants actually say rather than auditing the inputs.
What this does not tell you
- The robots.txt check is a snapshot taken on 27 July 2026, four days after the announcement. Yelp can change it at any time, and a licensing deal does not require a crawl path, so it may never change.
- Robots directives are instructions, not enforcement. They tell you what a well-behaved crawler is permitted to do, not what every system does.
- None of the coverage gives a rollout date, a volume, or a description of how OpenAI ranks or selects Yelp records inside an answer. Both companies declined to disclose financial terms.
- The review and revenue figures come from Yelp's FY2025 Form 10-K, so they are audited and dated 31 December 2025. They describe the corpus and the business, not what OpenAI actually receives under the contract.
- The deal is non-exclusive. Yelp is free to license the same content to other AI companies, so any advantage this creates for ChatGPT may be temporary.
Common questions
Can I get my Yelp page cited in ChatGPT by optimising it for AI crawlers?
No. Yelp's robots.txt blocks every AI crawler and agent from every business page. OAI-SearchBot, GPTBot, ChatGPT-User, ClaudeBot, Claude-User, PerplexityBot, CCBot, Google-Extended and Meta's crawlers are all disallowed from the site, with narrow exceptions for Yelp's own editorial articles. The catch-all rule for unnamed agents is also a full disallow. That was still true on 27 July 2026, four days after the OpenAI deal was announced. There is no crawl path to a /biz/ page, so nothing you write on your Yelp profile becomes visible to an assistant by being fetched. It becomes visible only if Yelp sends it through the licensed feed.
What did Yelp and OpenAI actually agree?
Yelp licensed its reviews, ratings, photos, and business information to OpenAI for use in ChatGPT, reported by Axios on 23 July 2026. Coverage puts the corpus at 330 million cumulative reviews and more than 8 million business listings. ChatGPT answers that use the content will carry Yelp branding and links, with OpenAI controlling how it is presented. Yelp also plans to bring its Request a Quote feature into ChatGPT so users can contact a business from the answer. Financial terms were not disclosed and the agreement is non-exclusive. Yelp shares rose about 8% on the news.
Are all 330 million Yelp reviews going to inform ChatGPT answers?
The 330 million figure is the count of reviews ever submitted, not the count that Yelp itself treats as reliable. Yelp's FY2025 10-K reports 330.2 million reviews submitted as of 31 December 2025, of which 300.3 million were available on business pages, including 48.1 million that its recommendation software does not recommend. Another 29.9 million had been removed. That leaves 252.2 million recommended reviews, 76% of the headline number, and those are the only ones Yelp lets factor into a star rating. Yelp has not said which slice OpenAI receives, so treat 330 million as the marketing number and 252 million as the one that carries a verdict.
Does this deal help hotels, vacation rentals, or coworking spaces?
Much less than it helps a plumber or a restaurant. Yelp reports its advertising revenue in two groupings: services, which is home, local, auto, professional, pets, events, real estate and financial services, and restaurants, retail and other. Lodging is not named in either grouping, so hotels, short-term rentals, and coworking sit in the residual. Services also produced 68% of Yelp's 2025 advertising revenue and grew 8% while the other grouping fell 6%, which is where Yelp's commercial attention and its freshest data sit. If you run a stay-based or space-based business, this is a small improvement to a source that was never strong for your category, not a new channel.
What should a local business actually do about it?
Fill in the Yelp profile as if it were a data feed, because it now is one. Claim the page, complete hours, service area, categories, licences, price range and photos, and fix anything wrong, because an error in an AI answer has no results page beside it to correct the record. Turn on Request a Quote and answer it. Work on steady, unprompted reviews from established Yelp accounts rather than review bursts, which is what the recommendation filter is built to catch. Then measure the answers themselves: ask the assistants the questions your customers ask and record what gets named and cited.
Why does a licensing deal matter more than crawlability for local search?
Because it changes who decides. When a source is crawlable, an assistant retrieves it, and you can influence what it reads by changing what you publish. When a source is licensed and closed, the assistant receives whatever the licensor chooses to send, ordered by the licensor's own logic. Yelp's recommendation software already decides which reviews count and which sit on a secondary page, and no employee, business owner, or reviewer can override it. That decision now runs upstream of an AI answer, and the only lever you have on it is your Yelp profile and your review history.
About the author
Ben Tannenbaum is the founder and CEO of Aiso, an AI search visibility platform that tracks how brands are cited in ChatGPT, Claude, Perplexity, Gemini, and Copilot. He studied statistics and previously founded a startup in commercial real estate. He writes on AI search for Search Engine Land and has published two papers on arXiv measuring AI-search demand: answer-reconstruction search density and request state across multi-turn conversations. Connect on LinkedIn.
Sources used in this article: Yelp Inc. FY2025 Form 10-K (filed 27 February 2026), yelp.com/robots.txt (retrieved 27 July 2026), Axios (23 July 2026), and Search Engine Land (23 July 2026). Every figure is linked to its source above.
Find out whether ChatGPT names your business
A licensed source is one you cannot audit from the outside. The only way to know what changed for your locations is to ask the assistants the questions your customers ask and record what they name and cite. Aiso does that on a schedule across ChatGPT, Claude, Gemini, Perplexity, and Copilot.