Google rebuilt how it answers "restaurants near me." Here's what changed.

On May 19 at its I/O conference, Google rebuilt AI Mode — the conversational layer that increasingly sits where the ten blue links used to be. A new underlying model, a redesigned search box (the first real change to that box in more than two decades), background agents that go research on their own, and one feature that matters more to a restaurant than all the rest: the answer can now book the table. AI Mode already reaches more than a billion people a month, and the rebuild rolls out across the US over the summer.

For an operator, the headline is not the model. It is that the local result stopped being a list and became a composition. Google no longer hands the diner ten links and a map and lets them sort it out. It reads the question, picks a few places, writes a short answer about them, and offers to act. The job changed from “rank in the list” to “be one of the few the model decides to name — and can act on.”

What “generative” changes about a restaurant result

A generated answer — the same surface as a Google AI Overview — is assembled from whatever the model can read and trust about you: your Google Business Profile, your reviews, the text on your own pages. It is not a ranked feed. The old map pack rewarded proximity and rating; the generated answer rewards legibility — can the model state, in a sentence it’s willing to put its name on, what you serve, when you’re open, and whether you fit the diner’s ask. If that sentence isn’t readable somewhere, you’re not in the paragraph. We walked the four signals that decide it in the four-number check.

The agentic part is the bigger shift

Generated answers have been creeping in for a year. The new thing at I/O is that the assistant can complete a task — and for restaurants, the task is the reservation. Through OpenTable, Resy, and Tock, AI Mode can book a table inside the answer, without sending the diner to your site at all. That is a gift if you’re connected to one of those providers and a quiet leak if you’re not: the diner asks for “a quiet spot for four on Friday,” the assistant finds one that can be booked, and books it. A restaurant that takes reservations by phone only — with no reservation link the agent can complete — is invisible to that flow no matter how good it is.

Three moves that keep you in the answer

None of this rewards a bigger ad budget. It rewards being readable and bookable — both of which you control.

Complete the profile. Hours by day, cuisine, attributes, the menu link — the fields the model quotes. Run the GBP grader to find the blanks.

Connect a booking provider. If you take reservations, get them into OpenTable, Resy, or Tock so the agent can finish the job. The how-to is in booking inside the AI answer.

Write the answer in text. Put the facts a diner asks for in plain text, backed by restaurant schema so the model can lift them, then check the whole front door with storefront health.

The rebuild is not a reason to panic. It is a reason to be legible. The restaurants that lose are the ones the model can’t read and can’t book — and both of those are fixable this week, without a contract or an ad spend.