Yelp Inc. integrated OpenAI's GPT-Live-1 voice model into Yelp Host and Hatch, betting that full-duplex conversation — software that handles interruptions and overlapping speech — converts restaurant reservation calls and service-business leads into booked revenue rather than just better-sounding demos.
"We're excited to be among the first to integrate GPT-Live-1 into Yelp Host and provide the leading AI voice solution for restaurants," Akhil Kuduvalli Ramesh, chief product officer at Yelp, said. "Powered by our reliable business data and frontier tech stack, Yelp Host's value to restaurants is not only how it sounds, but how deeply it understands each restaurant's operations."
The San Francisco company said Yelp Host and Hatch are among the first products to deploy GPT-Live-1, which OpenAI is also releasing through its API. Yelp Host has handled more than 1 million calls for restaurants since launching in October 2025. Hatch, acquired by Yelp in 2026, manages tens of millions of leads across voice, text, email and web, a book Yelp attributes to billions of dollars of annual value for its customers.
The architecture matters more than the voice quality. Full-duplex lets a caller interrupt, switch topics mid-sentence, or talk over background noise while the agent adapts, and Yelp says the system reads tone — excitement, frustration, impatience — and responds accordingly. Layering Yelp's own language handling on top of GPT-Live-1 also lets the agents detect and answer callers in nearly any language, closing a gap that has historically pushed non-English callers to voicemail.
Yelp's differentiator is the data underneath the model, not the model itself. Yelp Host draws on business information on Yelp, reservation availability, food ordering, menu specials and seating areas, so the agent can act on a specific restaurant's operations. Hatch runs on scheduling rules, technician availability, service areas and emergency routing. Chris Bache, co-founder and chief executive officer at Hatch, framed the split directly: "Great voice AI requires more than a great voice model. We've dedicated years to building the intelligence that understands how service businesses operate."
Early testing shows longer sentences, fewer transfers
Yelp reported two operational signals from production testing at scale: an improvement in call handling and a reduction in call transfers. Callers spoke in full, natural sentences rather than short voice commands, and post-call transcription accuracy improved, giving operators cleaner records of each conversation. Yelp did not disclose the size of the call-handling improvement, the transfer reduction, or the test conditions behind either figure.
The competitive field is crowded and the incumbents are not standing still. Toast, Square and Clover sell restaurant software with their own ordering and reservation layers, while ServiceTitan and Housecall Pro own scheduling for home-services contractors — the same operational data Hatch is built on. Alphabet's Google and Amazon's Alexa+ have pushed conversational voice into consumer and small-business workflows, and OpenAI's decision to ship GPT-Live-1 through its API means any competitor can buy the same front-end voice layer Yelp just adopted. Yelp's defense is the proprietary operational data, not exclusive access to the model.
That distinction sets the terms of the trade. Yelp shares have traded at a discount to software peers for years because the core advertising business grows slowly and the market has questioned whether AI features can be monetized. Voice AI gives Yelp a second revenue line — Yelp Host and Hatch are sold as subscriptions to businesses, not as free features — but the company disclosed no pricing, no attach rate and no revenue contribution from either product. The 1 million Yelp Host calls since October 2025 is a volume figure, not a revenue figure, and Yelp has not said how many of those calls converted into reservations or orders.
The next test is disclosure. Yelp's third-quarter results, expected in early November, will show whether voice AI shows up as a named line item or stays inside the services segment. Until then, the GPT-Live-1 integration is a product-positioning event with a measurable volume base and an unmeasured revenue base — the same gap that has kept the stock cheap.
This article is for informational purposes only and does not constitute investment advice.