Yelp's Quiet AI Year Is Suddenly Loud: Yelp Host Enters Live Tests

I was reading Yelp’s Q2 release on the train this morning, half-expecting another quarter of polite local-ads growth, when one line in the shareholder letter made me put down my coffee: Yelp Host is in live testing. Not a roadmap slide, not a beta sign-up — actually answering calls in real restaurants right now. That’s the disclosure that matters, and it makes Yelp the third platform this month to stake a claim on the restaurant phone.

The headline numbers were fine, almost deliberately so. Net revenue came in at $370 million, up 4% year over year. Net income of $44 million, up 16%. Adjusted EBITDA of $100 million at a 27% margin, up 10%. The company bought back $65.9 million of its own stock in the quarter. If you squinted, this was the kind of quarter a mature platform reports when nothing is happening. The shareholder letter tells a different story.

The line in the letter that changes the read

Tucked between the segment commentary and the buyback math, Yelp said Yelp Assistant project submissions were up roughly 400% year over year, and that Yelp Host — the AI call-answering product the company has been alluding to for two quarters — has entered live testing with restaurant operators. Read literally, that means Yelp is now answering phones, taking modifications, and routing reservations and to-go inquiries on behalf of real venues. Read between the lines, it means Yelp has decided the front-of-house phone is core surface area, not a curiosity.

I want to mark that as interpretation, not company language: Yelp didn’t frame Host as a platform pivot. They framed it as a feature that improves lead quality. But the practical effect of a third-party voice agent on a restaurant’s main line is that the AI becomes the first impression — for guests, for delivery dispatchers, for the press calling about a bad review. That’s a different relationship with the operator than running an ad auction.

The Assistant number deserves a separate beat. A 400% YoY lift in project submissions, as Yahoo Finance’s coverage of the call notes, tells you the consumer side of Yelp’s AI stack is compounding faster than the ad business it sits on top of. Assistants generate leads; Host answers the calls those leads place. The loop is finally closing inside one company.

Why “third this month” matters

Here’s the contrarian read: a single platform launching an AI phone agent is a product announcement. Three platforms doing it in the same month is a category forming, and the operators I talk to are about to spend the rest of Q3 figuring out which ringtone belongs to whom.

Vox AI’s seed round is coming later this month (forthcoming on the calendar I’m tracking), and Loman AI’s seed is forthcoming as well — both are pitching the same surface from the startup side. Whether or not those two land exactly as I expect, the directional bet doesn’t change: by Labor Day, a mid-size independent restaurant in a metro market will have three credible vendors knocking on the door offering to answer its phone. One of them owns the local search graph. The other two own dedicated voice stacks built from scratch.

For the platform-versus-pure-play debate, I’d point readers at a forthcoming May piece on the Yelp AI stack, which laid out how Assistant, Connect, and now Host are meant to interlock. And for the broader market frame, an upcoming May framework piece on voice agent maturity sketches the rubric I’m using to grade these launches: turn-taking latency, modification handling, escalation paths, and POS write-back. Host, on paper, has the platform-side advantage of already knowing the business. The pure-plays have the advantage of starting with the phone as the product, not as a tile in a feature grid.

What I’m watching next

Three things will tell us whether Host is a real product or a press line.

First, distribution. Yelp didn’t disclose the number of test restaurants or the geographies. If Host shows up as a paid add-on inside the existing Yelp for Business dashboard within ninety days, that’s a real launch. If it stays in invite-only purgatory through year-end, the live-testing language was aspirational.

Second, integration depth. Answering a phone is the easy part of an AI phone agent. The hard part is writing the reservation back to OpenTable or Resy, the to-go order back to Toast or Square, and the waitlist position back to the host stand tablet. I’ll be looking for partner names in the Q3 letter.

Third, the metric Yelp chooses to brag about. If it’s “calls answered,” that’s a vanity number. If it’s “reservations confirmed without human handoff” or “after-hours revenue captured,” Yelp is serious about owning the outcome, not the interaction.

The quarter itself, then, is the kind of thing a CFO loves and a beat reporter politely files. The Host disclosure is what I’d circle in red. A 4%-growth, 27%-margin local platform doesn’t usually decide to answer your phone — unless it has concluded the phone is where the next ten years of local commerce gets routed. On the evidence of this letter, Yelp has concluded exactly that, and it’s no longer the only one.

— Maya covers restaurant tech for TableTransfers. Tips: [email protected].