Yelp Moves Voice AI From Conversation to Conversion
Yelp and Hatch are among the first platforms integrating OpenAI’s GPT-Live-1, giving restaurant and service-business callers more natural conversations with AI. The resulting voice AI is connecting conversational models with the location data, business rules and operating systems required to turn a call into a booking.
Voice AI is getting better at sounding human. OpenAI’s new GPT-Live-1 can listen and respond simultaneously, accommodate interruptions and background conversations, and better maintain the natural rhythm of a call. Yelp has integrated the model into Yelp Host, its restaurant voice AI product, and Hatch, its AI communications platform for service businesses.
Those improvements could remove some of the friction consumers have traditionally encountered with automated phone systems, but they don’t necessarily solve the harder problem for restaurants, franchises and multi-location service brands: knowing what is happening at an individual location and being able to act on that information. Yelp’s architecture provides an example of how that distinction is emerging. GPT-Live-1 serves as the front-end voice layer interacting with the caller, while Yelp Host and Hatch retain their business data, operational intelligence and connections to the systems needed to do something with the customer’s request. The increasingly natural voice may be only one part of the value proposition; the more difficult layer could be everything behind it.
A Better Conversation Is Only the Beginning
GPT-Live-1 uses a full-duplex voice architecture, allowing the model to listen and speak simultaneously rather than relying on the more rigid back-and-forth exchanges associated with many automated voice systems. A caller can interrupt, pause, change direction or speak with someone in the background without necessarily disrupting the interaction.
That matters because actual customer calls rarely follow a clean question-and-answer sequence. Someone making a restaurant reservation might change the time after checking with another person, ask about outdoor seating midway through the conversation or pause while confirming the size of the party. A homeowner calling about a repair may describe a problem, correct the address and then ask about availability.
Yelp says early production testing of the new model has improved call handling and reduced transfers. The company has also seen callers using fuller, more natural sentences instead of the short commands people often use when they know they are interacting with an automated system. Those are early company-reported findings rather than independent performance benchmarks, but they provide one indication of how improved voice technology could change consumer behavior.
Yelp Host already has meaningful call volume behind it. Since launching in October 2025, the product has handled more than one million restaurant calls, according to Yelp, the equivalent of nearly two years of phone conversations. The company reported in July that call volume had grown an average of 38% month over month from launch through June 2026.
The Intelligence Behind the Voice
More natural conversation may make voice AI easier for consumers to use, but the technology still needs to know what to say and what it is allowed to do. “Great voice AI requires more than a great voice model,” Hatch CEO and co-founder Chris Bache said in announcing the integration. In a supporting post, Bache expanded on that distinction, arguing that service businesses require deterministic business logic alongside conversational AI.
A customer calling about an urgent repair, for example, may require the system to determine whether an address falls within the company’s service area, check real-time availability and apply the appropriate scheduling or emergency-routing rules. Hatch’s underlying intelligence is designed around those types of service-business requirements. Yelp Host applies the same general principle to restaurants, drawing on restaurant-specific information including business details, reservation availability, food ordering, menu specials and seating areas.
“Yelp Host’s value to restaurants is not only how it sounds, but how deeply it understands each restaurant’s operations,” Akhil Kuduvalli Ramesh, chief product officer at Yelp, said. For multi-location brands, that distinction becomes increasingly important because a sophisticated voice model can be available to many companies through the same API. Connecting that model with accurate location-level data, business rules and operating systems is a different challenge.
A natural conversation isn’t particularly valuable if the AI gives the wrong hours, doesn’t understand what a location offers, schedules outside a service territory or promises an appointment that can’t actually be fulfilled. As voice agents become customer-facing representatives of the brand, the operational context behind them becomes part of the customer experience.
From Call Handling to Conversion
Yelp Host isn’t limited to answering restaurant questions. It can handle reservations and takeout orders, giving the system the ability to move from customer inquiry toward a transaction. Hatch can qualify leads, check service areas and availability, schedule jobs and route calls that require human intervention. Together, those capabilities illustrate how voice AI can move from primarily being a labor-efficiency tool toward becoming part of a brand’s conversion infrastructure.
For multi-location marketers, the distinction is important. A paid search campaign, local listing, social post or AI recommendation may successfully generate consumer demand, but the customer journey can still break down when someone calls a location and nobody answers. Restaurants are particularly susceptible during busy service periods, while service businesses can miss high-intent leads after hours or when employees are already handling other customers.
Voice AI potentially closes some of that gap. Instead of measuring success primarily through calls answered or staff time saved, brands can begin evaluating whether those conversations produce reservations, orders, qualified leads and booked appointments. If a media investment generates an inbound call and an AI agent ultimately converts that caller into a customer, call handling is no longer simply a back-office operating function; it becomes part of the path from acquisition to revenue.
Scale Changes the Voice AI Equation
The opportunity becomes more complicated across a distributed organization. A single restaurant can train employees on how to answer common questions and manage reservations, while a restaurant chain operating hundreds of locations has to maintain that experience while accounting for differences in hours, menus, seating, availability and other local conditions. Service brands face similar variation in technician schedules, service territories and operating rules.
Voice AI offers a way to centralize parts of that experience without necessarily making every interaction generic, but that only works when the AI has reliable access to the underlying location data and operating systems. The challenge increasingly resembles the one brands already face elsewhere in AI-powered local discovery: centralized governance has to coexist with accurate local context.
Brands will need to determine which actions voice agents can complete autonomously, which require approval or escalation, how quickly location-level changes reach the AI and how performance is monitored across the network. Those questions become more consequential as voice agents take on tasks that directly affect revenue and customer experience. An incorrect answer from an automated FAQ is frustrating; an AI agent incorrectly booking a reservation or dispatching a service call creates a different level of operational risk.
Conversation and Conversion Begin to Converge
The announcement also fits into a broader change underway at Yelp. Street Fight recently examined Yelp’s integration of Reservations and Waitlist capabilities into ChatGPT, which allows consumers to move from an AI-driven restaurant recommendation toward taking action within the conversational experience. That development illustrated how recommendation and conversion are beginning to converge as AI takes on more of the traditional discovery journey.
Voice AI approaches the same shift from another direction. Instead of beginning with an AI recommendation and moving toward a transaction, Yelp Host and Hatch begin with an inbound customer conversation and attempt to turn it directly into an action. For a restaurant, “Do you have a table tonight?” can potentially end with a reservation; for a service brand, describing a broken air conditioner can end with a scheduled technician visit. Each removes steps between expressed customer intent and conversion.
As frontier voice models become available to more developers and platforms, access to the conversational technology itself may become less of a differentiator. For restaurant chains, franchises and multi-location service brands, the harder work sits behind the voice: maintaining accurate location data, availability, customer context and operating rules, then connecting those inputs to systems capable of completing an action. The voice may be what the customer experiences, but the intelligence behind it will determine whether the interaction actually works.
