Why Reservation Platforms Block AI Agents

The intuition most people arrive with is that booking platforms hate automation. That is not quite what the record shows. Across 2026 the same companies that restrict bots in their terms also shipped agent booking themselves, on purpose, with press releases. The restriction is real and the enthusiasm is real, and understanding how both can be true is the whole point of this page. Last verified 24 September 2026.

Do the platforms actually forbid AI agents?

They forbid a category, and the category is narrower than the word “agent” suggests. Resy stated to CNN Business on 23 September 2026 that it “does not currently permit unapproved third-party bots or agents to independently access or interact with the Resy platform.” Two qualifiers are doing the work there: unapproved, and independent. Meanwhile Resy went live as a booking surface inside ChatGPT on 10 August 2026, and Google named SevenRooms, TheFork, ResDiary and Mozrest as partners when AI Mode went agentic worldwide on 10 April 2026. An agent that arrives through a door the platform built is welcome. An agent that lets itself in is the problem.

Where is the line drawn in practice?

Four attributes separate tolerated automation from the kind that draws action, and none of them is “a program was involved”:

• Approval. A negotiated integration comes with rate limits, an identifiable client and someone to call. An unannounced client has none of those.
• Request volume. This is the measurable one, and it is what surfaces in logs long before anyone reads intent.
• Identity. One human, one account, one real name is routine. Rotating accounts and throwaway names are the signature of inventory farming.
• Follow-through. Tables claimed and never sat cost the restaurant directly, which is why no-show rates sit under so much of this enforcement.

Why is request volume the signal that matters most?

Because it is the only one visible instantly and without judgment. Availability lookups are not free for the platform: each one hits live inventory, and hard rooms are exactly where the most clients are watching at once. CNN reported one agent generating around 200 requests an hour continuously and polling every 0.4 seconds near a drop. Multiply a pattern like that by a few thousand ambitious users and you have a load profile shaped by software, aimed at the handful of restaurants that already draw the heaviest traffic. A platform does not need to decide whether that user is a nice person to decide it cannot serve that rate. The mechanics of why those particular rooms attract the pile-up are in why Resy tables vanish instantly.

What is the business reason underneath the rule?

Reservation inventory is the asset these companies sell to restaurants, and anything that copies, mirrors or replicates it erodes the product. OpenTable made that explicit on 16 April 2026 with a client-agreement change requiring OpenTable be the primary table-management system, with the stated rationale including blocking unauthorized third parties from replicating bookings elsewhere, as reported by Restaurant Dive and Restaurant Business. Consolidation raises the stakes further: DoorDash bought SevenRooms for $1.2 billion, and Amex moved Tock restaurants onto Resy on 11 August 2026, putting a combined 25,000-plus venues behind one book while Tock’s own consumer site is being wound down. Fewer, larger books make unsanctioned access a bigger deal, not a smaller one.

Does an approved integration solve the problem for diners?

Only partly, and the gaps are specific. The Resy path inside ChatGPT is US-only, waitlists there run through Yelp, and you cannot change or cancel a booking inside the chat; changes go back to the partner app. PCWorld’s Ben Patterson, testing the Claude connector Amex ships for Resy, reported that it surfaced times but he finished the booking on Resy’s own site. More fundamentally, every sanctioned integration reads what is on the shelf and books it. None of them sit and wait. At a restaurant showing nothing for eight weeks, an approved agent has nothing to book, which is a different limitation from a policy one. We cover that gap in booking hard tables with an AI assistant and in the OpenTable ChatGPT integration.

Where does SeatSwiper sit against this line?

Stated plainly, with no flattering adjectives: SeatSwiper is not an approved or endorsed partner of Resy, SevenRooms or OpenTable, and I am not going to imply otherwise or suggest it goes unnoticed. Terms restricting unapproved automation apply to it. What is also true, and is the part worth weighing, is that it is built against three of the four attributes above rather than around them. It books on your own account, in your own name, so there is no second identity. It books the one table you described, so nothing is claimed speculatively. It exists to get you into a seat rather than to accumulate inventory, which is the follow-through half. The fourth attribute, approval, it simply does not have, and you should factor that into your own decision rather than take reassurance from me. For the account-risk version of the same question see whether a reservation service can get your account banned and the reported incident in the Resy deactivation CNN covered.

Coverage is Resy, SevenRooms and OpenTable only, which leaves out restaurant-run booking pages, phone-only rooms and private dining. On price: the first successful booking is free, after that it is a flat $5 charged only when a booking actually happens, and nothing recurs. Starting a cancellation watch costs $1, and that dollar is billed when the watch begins even if no table ever opens; the free first booking does not absorb it. A watch that wins then costs the same $5. It can lose a race, and it will tell you when it does.