AI's New Gold Mine Is Trusted Data — and Yelp Is Sitting on a Rich Seam

The Trust Gap Behind AI's Search Boom

Artificial intelligence assistants can now recommend a restaurant, a doctor or a roofer in seconds, but usage is growing far faster than trust. According to Pew Research Center data cited in the report, 49 percent of US adults now use AI chatbots, up from 33 percent in 2024. Yet only 6 percent of people who see AI-generated search summaries say they trust them “a lot.”

Yelp’s own research points in the same direction. In its survey, 65 percent of respondents had used an AI-assisted search tool in the previous six months, but only 15 percent said they largely trusted the results, and 63 percent said they cross-checked AI answers against other sources.

That contradiction is the commercial backdrop for a notable deal: in July, OpenAI licensed Yelp’s ratings, reviews, photos and business information for use inside ChatGPT. Users can already make restaurant reservations or join waitlists through Yelp in ChatGPT, with requests for home-services price quotes planned next.

The wider lesson is that producing an answer has become cheap; proving that the answer is correct is becoming the expensive, strategically valuable part. Yelp, which was written off as an AI casualty a few years ago, now sits on data that AI companies are willing to license.

Why Yelp's 20-Year Review Archive Just Became an AI Asset

The Trust Deficit Is Driving Demand for Provenance

The usage data creates a specific incentive for AI builders. If almost half of US adults use chatbots but only a small minority trust AI search summaries, the product differentiator shifts from speed to evidence. A striking example comes from a Nature study cited in the piece: a smaller model called OpenScholar, built on 45 million scientific papers, beat GPT-4o on certain difficult tasks. GPT-4o’s citations to current literature were wrong 78 to 90 percent of the time, while OpenScholar reached source accuracy comparable to human experts.

That does not mean smaller models always beat larger ones, but it does mean curated, domain-specific data can close the gap with raw model scale.

Yelp’s Obsession With Review Quality Becomes a Licensing Advantage

Yelp’s asset is not simply 330 million reviews, 8.4 million business locations and roughly 500 million photos. Its real moat is the system it spent years building to evaluate whether a review is trustworthy. In 2025, 70 percent of submitted reviews were recommended, 17 percent were not recommended, 11 percent were removed and 2 percent were withdrawn. The company does not let employees manually override that system, and it argues that treating every review as equally valuable would make the platform easy to manipulate.

That curation cost creates a trade-off — some genuine reviews are filtered — but it also produces exactly the kind of cleaned, structured data a chatbot needs when it tells a user which plumber or restaurant to trust.

The Strategic Risk: Becoming a Back-End Supplier

The open question for Yelp is distribution. In the ChatGPT deal, OpenAI controls the consumer conversation and the transaction handoff. Yelp may be cited more often than rivals — a Yelp-funded study counted about 512,700 citations across ChatGPT, Gemini, Google AI Mode and Perplexity, 3.4 times the nearest home-services platform — but being the underlying answer source does not automatically keep the Yelp brand in front of users.

That is the central tension in AI data deals: the data owner gains reach and licensing revenue, while the assistant owner gains the relationship. The commercial value of Yelp’s archive is clear; the long-term brand value is less certain.

What the Trusted-Data Shift Means for AI Builders and Data Owners

For companies that own large, continuously updated datasets, the Yelp–OpenAI contract is a concrete precedent rather than a theory. The practical implications fall into four areas.

• Audit proprietary data before a partner does. Yelp turned 20 years of reviews, 8.4 million business locations and roughly 500 million photos into a licensing agreement with ChatGPT. Hospitals with clinical outcomes, banks with transaction histories and manufacturers with operational records are the categories the report explicitly names as next in line.
• Treat citation quality as a product requirement. With only 6 percent of US adults saying they trust AI search summaries a lot, and GPT-4o’s literature citations found to be wrong 78–90 percent of the time, verifiable sourcing is not a cosmetic feature. Teams building AI products should measure and publish source accuracy, not just answer fluency.
• Negotiate for the customer relationship, not just a content fee. In the ChatGPT integration, OpenAI keeps the user interaction while Yelp supplies data. Data owners entering similar deals should secure visible attribution, transaction completion or lead handoff — as Yelp is doing with reservations and, potentially, home-services quote requests.
• Watch the home-services pilot. OpenAI and Yelp have said quote requests will be added to the system. If that expansion works, it will signal that licensed local-business data can move from search summaries into actual commercial transactions, raising the value of verified review archives.