AI Agents Won't Fix Your OTA Problem. They'll Inherit It.

You already lost the direct booking fight once. You lost it to Booking.com, to Expedia, to a review site that charges you to appear above your own name in search results. You lost it slowly, over about fifteen years, one dependency at a time, until the commission check started feeling like rent.

Now there's a new conversation happening, about AI assistants that will plan trips and book rooms and tables on a guest's behalf. A lot of that conversation sounds like rescue. Like the machine is finally going to cut out the middleman and connect the traveler straight to the person who runs the boat, the kitchen, the six rooms above the old post office. That is not what's happening. What's happening is the middleman is learning to talk, and it's still the same middleman.

Here's the number that should sit with you for a minute. HOTREC, the European hotel association, put 2025 direct bookings at 51.3% and OTA bookings at 29.9%, continent-wide, across all hotels.[3] That sounds almost healthy. But that average includes chains with marketing budgets, loyalty programs, and direct-booking apps that cost more to build than most independent operators will spend on software in a decade. Cloudbeds looked specifically at independent hotels and found OTAs taking 63.4% of their bookings.[3] Nearly two out of three. The gap between the average and the independent reality is the whole story, and it's a story that predates any AI assistant by years.

It gets worse once you look at what those bookings are worth. Cloudbeds ran the numbers on 90 million bookings and found OTA cancellation rates of 21.8%, against 10.6% for direct.[8] That's not a rounding error. That's a guest who books three similar rooms across three platforms, picks one the morning of, and leaves you holding a cancellation for a Friday night you could have sold twice over. A direct booking is a commitment. An OTA booking is often just a placeholder in someone else's shopping cart.

Now put an AI assistant in front of that shopping cart.

Why would an AI agent behave any differently than an OTA?

It wouldn't, unless something forces it to. An agent that books on behalf of a guest is optimizing for the same thing an OTA search page optimizes for: the best match it can find fastest, from the inventory it can actually see. If the inventory it can see is OTA inventory, because that's what's structured, tagged, and fed to it in a format it can parse, then the agent becomes an OTA with better manners. It will still be comparison-shopping your room or your table against four others in the same radius. It just won't show you the comparison happening.

Look at what the big platforms are actually building, because the shape of the product tells you who it serves. Booking Holdings has a tool called Vayo, invite-only beta as of late October, that pulls your trip confirmations out of your email, watches for price changes, and searches flights and hotels, but when it's time to actually book, it sends you back to the supplier to finish.[2] Restaurant reservations inside Vayo route through OpenTable.[2] That's not independence. That's Booking Holdings building a dashboard that sits on top of the exact same distribution system, with itself still in the middle of the restaurant leg.

Expedia announced more than thirty new AI features in late October, across Expedia, Hotels.com, and Vrbo, including a natural-language search and an AI property assistant that answers guest questions using the property's own details and reviews.[1] Useful, maybe, if you're listed with them. It does not reduce the degree to which Expedia decides who gets seen. Google's AI Mode is reportedly moving toward hotel booking and price tracking directly inside search, and Meta's Muse can apparently fill out forms and complete purchases with a user's approval.[4][5] Every one of these tools is a front door. None of them is building a hallway that leads straight to you without walking past a shelf of competitors first.

And even the front door doesn't fully work yet. Hotel-Online reported on tests run in mid-October where Claude and ChatGPT agents tried to reach checkout at ten different hotel groups. They got to checkout at five or six of them. Neither agent completed an actual reservation, and access varied by which agent, which network, which bot-management system was running on the hotel's site that day.[9] So right now, AI assistant bookings for hotels are inconsistent at best, blocked at worst, even for the chains with engineering teams dedicated to this problem. For a guide outfit running a four-person crew and a shared inbox, the question of whether an agent can even see your availability calendar hasn't been asked yet, let alone answered.

Does getting booked by AI mean anything if you're invisible to it?

This is where the OTA concentration numbers stop being background noise and start being the actual threat. HospitalityNet's October analysis points out that Booking.com remains the dominant channel for independents specifically, and raises a real concern: if AI assistants lean on the same inventory, reviews, and paid placement that OTAs already control, the assistants will just reinforce that concentration instead of breaking it.[7][10] An agent has to get its information about your place from somewhere. If the only structured, reliable, machine-readable source of truth about your six rooms or your tasting menu or your dawn departure times lives inside an OTA's database, that's where the agent will go. You won't be cut out by malice. You'll be cut out by default, the same way a restaurant with no website gets left off a map it never asked to be on.

The honest caveat here: none of this is happening at scale yet. Booking Holdings said in August that traffic to them from large language models was significantly below 1% of room nights.[3] No major chain or OTA has published real numbers on agent-driven bookings.[3][10] This is still early enough that nobody, including the platforms spending hundreds of millions on it, actually knows how big agentic booking becomes or how fast. That's not a reason to relax. It's a reason to fix the foundation while the building is still small enough to move.

The fix isn't glamorous and it isn't new. It's the same thing that was true before any assistant showed up: your own booking page, your own calendar, your own menu and pricing and policies, published somewhere an agent can actually read them without having to go ask Booking.com first. A tasting menu with a locked-in prix fixe and a clear no-show policy, sitting on a page you control, is worth more to you in this coming shift than any partnership you'll never get invited to. The difference between surviving this and getting flattened by it is whether a machine, when it goes looking for the person who actually runs the thing, finds you, or finds the version of you that lives three clicks deep inside someone else's platform. Bookaible exists for operators who want to check which one they currently are.

The guest booking a dawn departure from a dock doesn't care whether an app or a person made the reservation. They care whether the boat leaves on time and whether the person running it knows their name. That part hasn't changed and isn't going to. What's changed is who gets to stand between you and that guest while the reservation gets made, and right now the people building that layer are the same people who've been taking a cut of your business for fifteen years. They just changed the interface.

You don't need a seat at Vayo's table or a line in Expedia's thirty new features. You need your own front door to be the one the machine walks through first.