ChatGPT can book a table now. Can it book yours?

Until this summer the whole goal of AI visibility for a restaurant was to get named. We have written about where that name comes from: OpenAI's July 2026 licensing deal with Yelp and the grounding behaviour that sits underneath ChatGPT's restaurant recommendations. The assumption behind all of it was that the assistant hands the diner a name and the diner does the rest.

On 10 August 2026 that assumption stopped holding.

Can ChatGPT book a table at my restaurant?

Only if your availability is already sitting in one of three systems.

Yelp announced on 10 August 2026 that diners can "book a table or join a restaurant waitlist on Yelp without leaving the chat," across "thousands of top restaurants across the U.S. and Canada." The restaurant-side product named is Yelp Guest Manager. Android Headlines reported the same launch with three booking partners: Yelp, OpenTable and Resy, with live availability read into the chat.

The mechanics follow from that. The assistant is reading open time slots out of partner systems. If your Saturday seven o'clock lives in a diary by the host stand, or behind a form on your own website that emails you, it is not in the pool the assistant can see. You can be named in the answer and still be the only place in the answer the diner cannot confirm.

Being named and being bookable are now two problems

This is the structural change, and it is worth sitting with before anyone sells you a fix.

A diner asks for Italian near Union Square. The assistant returns three places. Two of them come with a live time at 7:45pm and a confirm button. Yours comes with a phone number.

Nothing about your listing got worse. Your Yelp category is still right, your hours are still right, the local paper still put you in its best-of list. The answer still contains you. What changed is that the competitor beside you now completes the job in the same window, and the diner has to leave the chat to choose you.

That is a distribution problem, and distribution problems do not get solved by content work. No amount of schema markup, no rewritten opening paragraph, no llms.txt file puts a bookable time slot into a pool your restaurant is not in. Anyone who tells you otherwise is selling the tool they already had. We keep a longer list of that behaviour in GEO pitch red flags.

The honest version: this has a price tag

The obvious move is to tell you to join. We are not going to, because for a lot of the restaurants we talk to the arithmetic does not work.

These are commercial platforms with monthly software fees, and in OpenTable's case a charge per seated diner on top. OpenTable publishes the entry-level figure itself. Its Basic plan page states that after the trial the plan is "$149 per month plus a per cover fee," and answers the no-show question with "with OpenTable, you only pay for seated diners." That is the shape of the deal: a subscription, then a charge attached to each body in a chair.

We are not printing comparable numbers for Resy or Yelp Guest Manager, because neither publishes a rate card and the comparison articles that rank for the query contradict each other. One says Resy charges no per-cover fee at all. The next gives a per-cover figure for Resy. When secondary sources disagree that plainly, the number is not worth printing. Ask each platform for its current rate card in writing and run it against your own covers.

Run the numbers the boring way. Take your monthly cover count, apply the platform's fee structure, and subtract the covers you would have filled anyway through your existing channels, because a platform that converts a regular who would have phoned you is a cost with no revenue attached to it. For a twenty-seat place that runs full on a neighbourhood following, the answer is often to stay off. For a hundred-cover room with midweek gaps, the answer is usually different. Neither answer has anything to do with AI.

There is also a switching cost that does not show up on the rate card. Reservation platforms hold your guest history, your notes, your no-show records and your covers data. Moving between them is a project, not a settings change.

The integration cannot cancel or change anything

Worth stating plainly because it lands on your host stand rather than on a spreadsheet.

Android Headlines' coverage is explicit: "if you want to change or cancel a booking, the chatbot can't do that. Users must manage those changes directly on the partner platform's app or website."

So a booking made in ChatGPT is a booking made by someone who never opened the platform, does not have the app, and will call you when their plans move. The operational load of a chat-originated cover is higher than a cover booked by a diner who was already inside OpenTable. Factor a share of extra phone traffic into the decision, and brief whoever answers the phone that these callers will not know which platform they used.

Hotels and tour operators: the same question, not the same answer

The split between being named and being transacted is structural, and it will reach every category where an assistant can reach live inventory.

We are not going to claim ChatGPT books hotel rooms or tours today. We have not seen an announcement extending this flow beyond restaurants in the US and Canada, and we are not going to infer one from a restaurant launch. What is worth doing now is the audit, because it costs an afternoon and the answer is useful either way.

Ask one question: if an assistant wanted to complete this transaction on a customer's behalf, where would the live availability come from? For a hotel that is your channel manager and whichever OTAs you already sit on. For a tour operator it is your booking engine and the marketplaces you list on. If the honest answer is "an enquiry form that reaches my inbox," you already know which side of the line you are on, whenever the rail arrives. The attribute work that gets you named is still the prerequisite, and we have written that up for hotels and for tour operators.

What to do in the next week

Three things, in order.

Check whether you are already reachable. If you run Yelp Guest Manager, OpenTable or Resy today, open ChatGPT, ask for a table at your own restaurant on a night you have availability, and see whether a time appears. That is a five-minute test and it tells you more than any dashboard.

If you are on none of the three, do the cover arithmetic before you do anything else. Get the rate cards in writing. The decision is about distribution economics and you should be able to defend it in one sentence either way.

Whatever you decide, keep the listing work going, because it is upstream of both outcomes. Being bookable does not help if the assistant never names you, and the name still comes from the listing surfaces and the pages other people write about you. That order of operations is in how restaurants get recommended by AI.

The one thing to avoid is treating a reservation platform as an AI visibility purchase. It is a channel with a commission, the same as it was in 2019. The only thing that changed in August is that the channel now has a second entrance, and the diner walks through it without leaving the chat.