The OpenTable ChatGPT Integration, Explained

Yes, OpenTable and ChatGPT are connected, and they have been in one form or another since the very first wave of ChatGPT plugins in March 2023. As of September 2026 the practical version looks like this: you can ask ChatGPT where to eat, get real OpenTable restaurants back in the conversation, and move from chat into an actual OpenTable booking without ever opening the OpenTable app yourself. For an ordinary Tuesday dinner at a restaurant with open tables, that flow genuinely works, and it is a nicer way to browse than paging through a list of thumbnails.

Here is the part the announcement posts skip: the integration changes how you find a table, not whether one exists. ChatGPT talks to the same OpenTable inventory you see in the app, at the moment you ask. If the restaurant you want shows nothing for Saturday, the assistant sees the same nothing, says so more politely, and suggests somewhere else. The integration is a discovery layer, not a line-skip. This page walks through the history of the tie-up, what booking through ChatGPT actually does mechanically, whose name ends up on the reservation, and where the whole in-chat model runs out of road, which is exactly the point where hard tables live.

The timeline, because "integration" has meant four different things

OpenTable has been unusually early to every OpenAI surface, which makes sense for a Booking Holdings company whose entire business is being wherever diners decide where to eat. The word integration has covered four distinct products, and knowing which era an article is describing saves a lot of confusion.

• March 2023, plugins. OpenTable was among the first third-party ChatGPT plugins, alongside Expedia and Zapier. You could ask for a restaurant and get handed an OpenTable link with your party size and date prefilled. Plugins as a product were wound down in 2024.
• January 2025, Operator. OpenAI launched its browsing agent with OpenTable as a named launch partner. Operator drove a real browser, so it could click through an actual booking flow on your behalf rather than just handing you a link.
• July 2025, ChatGPT agent. Operator was folded into a general agent mode inside ChatGPT itself. Same idea, one product: the assistant can browse, fill forms, and complete a booking on a site like OpenTable as part of a longer task.
• October 2025, apps in ChatGPT. OpenAI announced an SDK that lets companies run interactive apps inside the conversation, and OpenTable was named among the partners slated for it. This is the deepest version: structured restaurant results and booking steps rendered natively in chat rather than an agent puppeting a website. Rollout has been gradual and varies by region and account, so check the app directory inside ChatGPT for the current state rather than trusting any article, this one included.

Notice the direction of travel across those four eras: from a link, to a browser being driven for you, to the booking flow living inside the chat. What never changed is the thing underneath, which is a normal OpenTable reservation placed against whatever the restaurant currently has online.

What booking through ChatGPT actually does

Mechanically, every version of the integration is diner-side. ChatGPT queries OpenTable availability the same way the OpenTable app does, presents what it finds, and if you confirm, a standard reservation is created against the restaurant's live book. The restaurant does not see anything special. There is no partner inventory, no held-back allocation for AI bookings, no priority lane. OpenTable lists more than 55,000 restaurants, and for the overwhelming majority of them, the ones with real availability most nights, this is all you need. Ask, pick, confirm, done.

The reservation itself lands where you would expect: on the OpenTable account the integration is connected to, under that account's name, with the usual confirmation email and the usual cancellation policy attached. That own-account detail matters more than it sounds. A reservation made on your account in your name is one you can modify, cancel, and actually be seated against, and it keeps you inside every platform's rules. The broader question of trusting an assistant with your bookings gets a full treatment in whether it is safe to let AI book restaurants.

One restaurant-side wrinkle worth knowing. Restaurants pay OpenTable per seated cover, and how a diner arrives at the booking can affect what the restaurant is charged, which is why some venues push their own website's booking widget over the marketplace. A reservation initiated through ChatGPT is, from the restaurant's console, an OpenTable network booking like any other. None of that changes anything for you at the table, but it explains why a restaurant might quietly prefer that you book direct, and why the same room sometimes shows slightly different availability on its own site than in the marketplace or in chat. When the assistant says a restaurant has nothing, checking the venue's own booking page takes ten seconds and occasionally contradicts it.

Where the in-chat model runs out of road

The integration answers "what is available right now that suits me," and it answers it well. The question it structurally cannot answer is "get me the table that is not available right now." Those are different problems, and the second one is the only problem at the restaurants people actually fight over.

A hard table is, by definition, not on the book at the moment you ask. On OpenTable most venues run a rolling booking window, so new dates enter the calendar day by day, often overnight, and the desirable slots at a contested room are taken almost as soon as they appear (the mechanics are covered in when OpenTable releases reservations). After that, availability at a place like that exists only in brief flashes: a new date entering the window, or a cancellation putting a table back at some random hour. An assistant that queries availability at ask-time will hit those flashes only by luck. Ask ChatGPT for Saturday at a genuinely hard room and the honest transcript is: it checked, there was nothing, it offered alternatives. The model performed perfectly. The result is still no table.

Could agent mode just keep checking? Not usefully. A chat session is episodic: it runs while your request runs, then it ends. Winning a cancellation means watching one restaurant's book continuously for days and completing a booking within seconds of a table appearing, frequently at 11pm or 2pm on a weekday when no human, and no chat session, is active. That is a standing background process with your booking credentials, not a conversation. The same gap applies across every assistant, not just ChatGPT, and the fuller comparison lives in booking reservations with AI assistants.

Discovery layer and watcher are complements, not rivals

The clean way to think about it: the ChatGPT integration is a front end on present availability, and a watcher is an agent against future availability. The first is conversational and instant and works on the easy 95 percent of bookings. The second is silent and persistent and exists for the hard 5 percent. Neither substitutes for the other, and the failure mode to avoid is expecting the front end to do the watcher's job, which is how people end up asking an assistant for Saturday at a contested room every morning and concluding AI booking does not work.

The watcher half of the setup

SeatSwiper is that second half. It watches the specific table you want on OpenTable, Resy, or SevenRooms around the clock, and the moment availability appears, at a release or from a cancellation, it books it on your own account, in your own name, exactly as if you had been the one person staring at the page at the right second. First booking is free, then a flat $5 only when it actually wins you a seat. It does not support Tock. The full mechanics are in how SeatSwiper works. Used together the division of labor is natural: let ChatGPT handle "where should we eat Thursday," and put a watch on the room that shows nothing, because nothing at ask-time is not the same as nothing ever.