In this article
On 10 August 2026, ChatGPT quietly added a new button in the US and Canada: book a restaurant table without leaving the chat, through OpenTable, Resy or Yelp. No clicking through to a website, no phone call — the user asks, and the conversation handles the rest.
Google's Gemini has been doing something similar for a while through "Reserve with Google" (also on OpenTable), and picked up its own direct OpenTable connection this year. Perplexity, Microsoft Copilot and Amazon's Alexa+ each have their own version. Four years ago, "AI in hospitality" mostly meant a chatbot on your website or a pricing algorithm. This is something else: for the first time, an assistant completes the whole transaction — search, choose, book — without the guest ever opening your website or picking up a phone.
That isn't speculation about 2030. It's happening this month, at the two largest chat assistants in the world. And the question that comes with it isn't "should I do something about AI". It's the simpler, more urgent one: does your restaurant exist to one of these assistants, and what happens when it tries to book you?
This article answers that honestly — including the part that doesn't work yet. It picks up where AI Search: 7 Steps to Get Your Restaurant Found leaves off: that piece is about getting named in an AI answer. This one is about the step after — what an assistant needs to actually book that name.
The old path and the new path
Two ways a table gets booked today — and the difference isn't the technology, it's who makes the judgment calls along the way.
Today, the ordinary way
Today, via an AI assistant
On the old path, a human makes at least two judgment calls using context that's written down nowhere — "I can hear it's a birthday, I'll seat them by the window". On the new path, that context only exists if it's published somewhere. That's the thread running through this entire article.
The 4 things that decide if you're found and booked
Under the hood, this runs on a handful of new technical standards with names like MCP (Model Context Protocol, open-sourced by Anthropic in late 2024) and ACP (Agentic Commerce Protocol, from OpenAI and Stripe). You don't need to know those names — only what they do: they let an assistant read your availability and terms the way it reads a webpage today, and then act on them itself. Four things decide what that action turns out to be.
1. You have to be somewhere an agent can actually read you
Today, every assistant only talks to a handful of partners — and that list is shorter than you'd think. None of the big booking platforms (OpenTable, Resy, SevenRooms, Tock) opens its write API to just any third party; a connection runs through a formal partnership, not an open standard anyone can tap into.
For an independent venue, that's an uncomfortable truth: OpenTable, Resy and Yelp are strongest in the US and Canada, and that's exactly where the ChatGPT integration is confined for now. But the strategic lesson holds regardless of which specific platform ends up winning in your market. An assistant can only book what it can read — live availability, a written cancellation policy, current opening hours. A website with nothing but a phone number and a PDF menu has nothing to offer there, whichever platform knocks next.
Below is who talks to whom today, and where — so you know exactly which door is already open to you, and which one isn't yet.
The picture as of this writing (August 2026) — this shifts fast, but the pattern is stable: every assistant talks to a small, fixed set of platforms, never to "the internet".
ChatGPT
OpenTable · Resy · Yelp
US and CanadaDirectly bookable in the chat since 10 August 2026, including waitlists.
Gemini
OpenTable (Reserve with Google)
Wherever Reserve with Google is liveAlready existed via Google Search/Maps; got its own direct OpenTable connection in 2026.
Perplexity
Its own partner connections
Region-dependentAdds booking buttons to search results through its own partners.
Microsoft Copilot
Its own partner connections
Region-dependentFollows the same partnership model as the other assistants.
Amazon Alexa+
Its own partner connections
Region-dependentBooks by voice, under the same platform limits as the text assistants.
None of these five reads a random restaurant website directly and books straight off it. They read the platforms they have a formal connection with. If you aren't on one of them, your website is still perfectly readable to a human — to this agent, you simply don't exist yet.
2. Sometimes the caller is already an AI
There's a second direction this is already running in: OpenTable's own network of voice-AI partners — Loman AI, Maple, Hostie AI, Anvevo — already answers the phone for restaurants today and writes the booking straight into the reservation book. That isn't a pitch-deck promise; it's running for existing customers right now.
Practically, that means two things. If you're considering one of these systems yourself, the question is the same one this whole article keeps coming back to: does it have access to your genuinely current availability and policy, or is it just repeating whatever was set up three months ago? And if you aren't using one (yet), it means some of the calls coming in — from guests, or soon from their own assistants — meet someone who expects no small talk and picks up on no implied cues. Whoever answers the phone can just answer a literal question literally and completely; that's more a matter of instruction than of technology.
3. An agent doesn't ask twice — it reads, or it drops off
A human caller who gets a vague answer just asks again: "and what if we're not that many after all?" An agent that hits a page saying "call us for details" doesn't — it marks the field unknown and moves on, or stops. Four things come up most often, and this blog happens to already have a dedicated, deeper article for each one:
Deposit and cancellation terms. Is there a figure and a timeframe, or just "a no-show fee may apply"? See Deposit & Cancellation Policy: 7 Rules Against No-Shows for how to word it concretely.
Allergen and dietary information. An agent booking for a guest with a nut allergy searches for an explicit answer — not the absence of a complaint. See the allergen management article.
Group minimums and policy. From how many guests does a minimum spend apply, and is that written down anywhere? See Minimum Spend on Group Bookings.
Opening hours, including the exceptions. Not just "open noon to 10pm", but also the closing day, the renovation break, the public holiday that happens to fall on a Tuesday this year. See Optimizing Your Opening Hours.
4. What lands in your book is a stranger, twice over
A guest who calls you themselves gives you two things for free: a voice you can read something into (nervous, upbeat, a regular you already recognise), and a moment to ask something back — "is there still room on the terrace?". An AI-placed booking gives you neither. The name arrives without anyone at your venue having spoken to anyone, and without the history you'd already have on a regular.
That makes two things more important, not less: still actively confirm large parties (see Reduce Restaurant No-Shows: 7 Proven Strategies for how), and keep track of who your regulars are so an AI-booked name is recognised as a regular next time instead of forever treated as a stranger — exactly what a tool like tracking your regulars is for.
Test whether you're ready for an AI booking
Eight questions, none of them complicated — and the reason each one is here is one of the four points above. Tick what already holds true today. What's left isn't "something to get to eventually" — it's exactly what an assistant would already run into if it tried.
AI-booking readiness score
Tick what already holds true for your venue today.
—
These eight weights are an estimate based on how heavily each one matters to the four reasons above why a booking does or doesn't go through — not a scientific measurement. The point is a ranking, not an exact truth.
What doesn't hold up yet
Honesty belongs in this topic, because the marketing around "AI agents" constantly overstates it. What doesn't exist today: an assistant that randomly crawls the internet and books on any restaurant website it finds. What does exist is a small number of formal partnerships between a handful of chat assistants and a handful of booking platforms — for now mostly US/Canadian, and for now without an open, self-serve write API any venue can hook into on its own.
Changes and cancellations on an AI-placed booking still have to go through the platform itself today, not through the conversation that created them — an assistant books, but doesn't yet manage. And which platform ends up playing OpenTable's role in Belgium or the Netherlands is anything but settled.
What doesn't wait on who wins that fight, though, is the underlying question this whole article is about: is your availability, your policy and your contact information written somewhere a machine can read it, or only somewhere a human can interpret it? That's exactly what the score above measures, and it's worth doing today regardless of which assistant knocks tomorrow.
Conclusion: this isn't a future scenario any more
Four years ago, "AI and reservations" was a chapter in a vision statement. Since 10 August 2026, it's a button in ChatGPT that people are actually using. The next move for an independent venue isn't to jump onto every new platform — it's to get the groundwork in place that every one of those platforms needs, whoever wins: visible availability, written policy, a phone that gets answered, and a system that recognises a regular as a regular.
That, as it happens, is exactly what owning your own reservation system gives you instead of renting one — data that's yours, in a shape every next wave of technology can read, without re-integrating from scratch each time someone new gets big.