Yelp Is Shipping AI That Calls AI — and the Market Still Prices It Below 1x Sales

Yelp launched Yelp Host, an AI phone answering service for restaurants, in October 2025. By its own count the product had handled more than 1 million calls from launch through June 2026, and more than 1.5 million calls as of October 1, 2026. Yelp rebuilt the voice front end on OpenAI's GPT-Live-1 in September 2026.

Check for Me then dials those restaurants from the consumer side. The most likely default state of that call, on Yelp's own installed base, is an agent talking to another agent.

The volume curve is the part most readers will miss. Yelp Host averaged roughly 111,000 calls a month across its first nine months. Then calls crossed from 1.0 million to 1.5 million between June and October 1 — about 167,000 a month in the final quarter, roughly 50% above the prior run rate.

That inflection is the whole commercial argument for Check for Me. Yelp did not need consumers to adopt an agent so much as it needed restaurants to already be running one when the consumer's agent arrived.

Yelp reports advertising and everything else separately, and the gap between them has become the story. In Q2 2026 total net revenue rose 1.4% year over year to $375.5 million. Other revenue — which carries transactions, subscriptions and data licensing, and is where Yelp Host, Guest Manager and Request a Quote live — rose 98% to a record $33 million.

In FY2025 the same line grew 17%. One quarter of 98% growth does not make a trend. But it lands on top of a business that is buying its way into that line.

Yelp Host is sold as a subscription starting at $249 per month, with tiers running to $599 and an alternative per-conversation option at $0.59. At that listed conversation price, a restaurant needs roughly 420 calls a month to cover the entry tier — so the cost of the voice model is not what sets the economics.

What sets the economics is whether a restaurant renews at $249 a month. Check for Me is the demand-side reason to say yes: the agent that calls is the agent that fills the table and answers the phone Yelp is already selling.

In2026 Yelp signed four separate arrangements with OpenAI, and three of them pull in opposite directions. The sequence matters more than any single announcement, because it ends with Yelp shipping a consumer agent into a product that already answers local questions better than Yelp's own app.

Read together, the last three rows are a defensive perimeter. Yelp put its reservation and waitlist inventory inside ChatGPT in August, which surrendered the discovery surface but kept the transaction. Then two months later it built a consumer agent for the one task — live wait times on a walk-in — that a chat assistant still cannot do, because it requires placing a call.

The wait-time call is the wedge ChatGPT structurally cannot take, which makes it a narrow, defensible feature rather than a platform.

Check for Me lands on top of an industry that has already decided table inventory is a utility. On August 10, 2026 — the same day — Yelp Reservations and Waitlist went live inside ChatGPT, alongside Resy. OpenTable, the reservation unit of Booking Holdings, is now reachable both through Yelp Assistant and through ChatGPT. Resy, owned by American Express, launched in ChatGPT the same week.

A diner can now book the same table through three interfaces. None of them owns the restaurant relationship; all of them rent it.

The competitive pressure shows up in Toast's numbers, not Yelp's. Toast reported Q2 2026 revenue of $1.91 billion, up 23.1%, with locations up 22% to roughly 180,000 and annualized recurring revenue up 25% to $2.4 billion. Against that, Yelp's advertising revenue from restaurants, retail and other categories fell 6% in FY2025 to $444 million.

Yelp's agent products are a defense of a shrinking restaurant ad base, not an independent growth story, and the pricing — $30 a month, $249 a month — is set against a platform already billing $2.4 billion of ARR.

There are two credible ways to build a voice-agent stack in 2026, and the public market is pricing them as if they were unrelated businesses. One rents a frontier model and owns the distribution. The other sells the model and owns nothing downstream. The first is growing a services line inside a marketplace; the second is a standalone infrastructure story at a multiple that assumes the growth never stops.

SoundHound AI is the cleanest listed read on the voice-agent category: revenue up 45% to $61.9 million in Q2 2026, full-year guidance raised to $230 million–$260 million, and a market value near 9.2x trailing sales while burning cash. ElevenLabs, private, went from $330 million of ARR at the end of 2025 to roughly $600 million by June 2026 and was valued at $22 billion in a September 2026 employee tender.

Yelp grows that layer faster and collects cash doing it — yet trades below one times sales. The gap is the trade.

The governing precedent is old but narrow. In a declaratory ruling adopted February 2, 2024 and released February 8, 2024, the FCC held that the TCPA's restrictions on an \"artificial or prerecorded voice\" encompass AI technologies that generate human voices, and that such calls require the prior express consent of the called party.

Check for Me calls a restaurant's published business line and then texts the consumer who supplied a phone number. Yelp disclosed no consent architecture, no disclosure language, or telco arrangement with the release. The compliance perimeter here is legally unsettled and company-undisclosed, and that gap will be settled by whoever gets challenged first, not by prior art.

Trailing net income is $126.5 million and trailing free cash flow is $296.8 million, and Yelp has repurchased $349.2 million of stock over the trailing twelve months — roughly 33% of its current market value, retired in twelve months. Cash stands at $94.3 million after funding the Hatch purchase.

Against that, guidance for 2026 calls for adjusted EBITDA of $310 million to $330 million on essentially flat revenue. A market value near $1.1 billion prices the business at roughly 3.2x the midpoint of that guidance and near 3.7x trailing free cash flow, with shares down from a 52-week high of $34.49 to under $20.

What the market is pricing, plainly, is a search-disintermediation story: restaurants advertising down 6%, total revenue up 1.4%, and a top line that has sat between $358 million and $376 million for six straight quarters. On that reading the AI work is defensive spending on a shrinking base.

The bull case requires something different — that owning the restaurant's phone line, the waitlist and the consumer app is worth more than owning the reservation dialog, and that closed-loop agents become the default way local services are reached.

Strip away the novelty and the fact pattern is this: one listed company sits simultaneously at the consumer end of a phone call, the business end of the same phone call, and the data layer that answers the question the call produces. OpenTable and Resy own table inventory but not the caller. ChatGPT owns the assistant but must borrow both the inventory and the answer.

The honest caveat is that this advantage is worth $33 million a quarter today, inside a company with a flat $1.5 billion top line. Check for Me is a well-chosen wedge with no disclosed monetization attached to it. The agent stack is the option, not the earnings — and at 0.7x trailing sales with a third of the market cap retired in a year, the market is currently paying nothing for it.