The OTA tax is coming for your AI bookings.

A traveler opens ChatGPT in May 2026 and asks for a small hotel in Lisbon with character, walking distance to good wine bars. ChatGPT recommends three properties through the Booking.com app that shipped at OpenAI’s DevDay in October 2025. She picks one, taps through to booking.com to confirm, and the reservation lands as a normal Booking.com booking.

The hotel pays the same fifteen to twenty five percent it has been paying Booking for a decade, depending on its Genius and Preferred Partner setup, on a channel that did not exist eighteen months ago. Booking gets the guest contact details and the right to retarget her for every trip she takes for the next ten years.

This is what AI hotel booking looks like in May 2026. If no hotel changes anything between now and 2030, this is what AI hotel booking looks like in 2036 as well.

The search layer has already moved

Travelers do not open Booking.com to find hotels anymore. They ask ChatGPT, Claude, Gemini, Perplexity, and a growing list of AI trip planners like Mindtrip, GuideGeek, and Layla that did not exist eighteen months ago. ChatGPT alone is now used by hundreds of millions of people each week, and the long tail of trip planners is still small but growing fast inside specific traveler segments. The thing OTAs spent a decade winning, which is the attention and the search step, has moved to a new set of surfaces that the OTAs do not own.

What the OTAs still have is the booking transaction itself, the brand trust at the moment of payment, the customer service operation, the multi-currency payments rails, and the loyalty wedge of Genius and One Key. That is real and it is not going to evaporate. But it is a much smaller piece of the value than it used to be, and it shrinks every quarter that travelers spend their planning time somewhere other than booking.com.

The OTAs see exactly what you see, and they are racing for the booking layer

The OTAs are not standing still. They are sprinting to insert themselves between the AI planner and the room.

• Booking.com is already inside ChatGPT. The integration shipped via the OpenAI Apps SDK at DevDay in October 2025. Hotel inventory is rendered inside the chat, the user is handed off to booking.com to confirm the reservation, and the commission to the hotel is unchanged. OpenAI walked back in-chat checkout in March 2026, which means the discovery-and-handoff pattern is now the explicit design rather than a transitional step.
• Mindtrip, one of the most established consumer AI trip planners, currently books hotels through Priceline.
• Every other major OTA has a clear commercial reason to ship equivalent placements before its competitors do, and the AI planners themselves are not yet locked into any single backend.

The pattern is the same one OTAs have used at every prior platform shift. Let the new surface do the inspiration and the search work, then show up at the moment of booking with the inventory and the payment rails already in place, and collect the same fifteen to twenty five percent that has been the OTA standard since the last shift. The interface changes, but the share of every booking that leaves the hotel stays roughly the same.

If hotels do nothing between now and the end of 2027, the OTA-routed integration is what AI hotel distribution becomes by default. The same OTAs that hotels have spent a decade trying to escape will collect commission on bookings that started in chat sessions the hotel never saw, on travelers whose preferences the hotel never heard.

The window closes around the end of 2027

The booking layer can still go a different direction. The window is open today because the AI planners themselves have not yet decided how to handle the booking step in the long run. Most of them would prefer a direct path to hotels for the obvious reasons, which are better unit economics on the share they earn, cleaner traveler data, real preferences they can act on inside the conversation, and no parity headaches when surfacing live availability. They are not married to the OTAs yet.

That openness has a shelf life. Booking shipped into ChatGPT in October 2025, OpenAI confirmed in March 2026 that checkout itself stays on the partner’s site, and the rest of the major OTAs have every reason to ship equivalent discovery placements through 2026. Once those placements exist and the AI planners have built their listing surfaces, their support escalation, and their refund handling against the OTA APIs, the routing is very hard to renegotiate, because the AI planner now has a working integration it would have to tear out and rebuild. Best read on the timing is that the patterns harden through 2026 and 2027, and by some point in 2028 the AI channel looks structurally like the OTA-dominated channel it grew out of.

If a hotel waits for the dust to settle, it ends up back inside the OTA system. It pays the OTA commission on a channel that did not exist when it started paying it.

It is not just about commission

The commission savings are the part of the story that looks obvious on the P&L. The more interesting part is what becomes possible when a hotel can talk directly to a traveler’s AI assistant before the guest arrives, and keep talking to it during the stay and after it.

• Preferences delivered in a structured form the PMS can actually route. “I sleep hot, I work late, I always want sparkling water” reaches the property as fields the front desk and housekeeping can read, instead of a free-text note buried in the OTA confirmation that nobody opens. Acting on those preferences is still a real piece of operational work inside the property, but the channel at least delivers them in a usable form rather than throwing them away at the handoff.
• A two-way conversation with the AI. The hotel can push late-cancellation availability, last-minute upsells, and unpublished rates back into live AI searches in the moments those rates are actually valuable. The OTA channel makes this operationally awkward today, partly because of parity contracts that still constrain most markets even after the European DMA rulings weakened them in 2024 and 2025, and partly because surfacing unpublished rates inside an OTA listing trains travelers to look outside the OTA, which is not a habit the OTA wants to build.
• Real guest contact details on every booking. The hotel sees the traveler’s real email, real phone number, and the preferences the traveler chose to share, instead of a booking.com proxy email and a generic OTA form. Strait does not sit between the hotel and the guest after the booking is made, and it does not store traveler profiles on the hotel’s behalf, which means the relationship the hotel builds from that first stay onwards is the hotel’s to keep.
• Direct economics through the stay and after it. Cleaner cancellations, fewer rate rules to manage, and the real ability to bring the guest back through the hotel’s own channels without paying a second commission on the return visit.

This is the relationship that loyalty programs have been trying to manufacture inside the OTA-dominated channel for the last fifteen years. The new channel can deliver it directly, if hotels take the position before the OTAs absorb it.

What hotels should do now

The right move is to get connected to the AI trip planners through a direct integration that lives on top of the PMS, rather than through an OTA wrapper, a bedbank reseller, or a channel manager built for the OTA era. A channel manager is the right tool for routing the same room across twenty OTA listings under parity. It is not the right tool for a channel where the agent on the other side is having a live conversation with the traveler about preferences, late availability, and unpublished rates, because the channel manager’s data model collapses all of that back into the OTA-style room and rate plan.

Strait sits next to the channel manager rather than replacing it. The channel manager keeps doing what it does on Booking, Expedia, and the other OTAs. Strait carries the AI planner traffic on a separate path that preserves the structured preferences, the live two-way availability, and the real guest contact details.

The commercial terms are simple. Seven percent commission, paid only after the guest checks out, with no setup fee, no exclusivity, and the ability to cancel at any time. The hotel is the merchant of record on the booking, and the consumer always sees the hotel’s PMS rate without a markup. Volume through the AI channel is still small in May 2026, smaller than what most hotels see from meta in a given month, and the honest reason to connect now is not the bookings the hotel will get this quarter. The reason is the position the hotel will hold in the channel when the volume arrives through 2027 and 2028, and the difference between holding that position directly and paying the OTA commission on it for the next thirty years.

Strait is the channel we are building for that traffic. We would rather hotels build it with us through 2026 and 2027 than wake up in 2030 wishing they had.