“AI only drives 1% of our bookings, so we’ll wait and see.” I hear some version of that in almost every travel boardroom right now. It sounds careful. It is actually the riskiest plan on the table, because the rules for AI booking are being written this year, and they are being written without most hotels in the room.
That is the real message in Skift’s look at AI travel booking and 2027 commerce budgets, published on 19 October 2026. The volume is tiny. The decisions are not.
The small number everyone is hiding behind
Let’s be fair to the “wait” camp. The numbers really are small. Booking Holdings has told investors that large language models account for well under 1% of its room nights. Bernstein’s Richard Clarke puts AI referrals below 2% at Booking and Marriott.
But budgets based on today’s volume answer the wrong question. The question isn’t “how much does AI sell for us now?” It’s “on whose terms will AI sell for us later?” I made a similar point about traffic when AI was sending hotels less than 1% of their visits. The small number is a measuring problem. This time it is a negotiating problem.
Airlines signed. Hotels got shopped.
Here is the example that should keep hotel commercial teams up at night. Skift tested Meta’s Muse agent. When it booked a flight, it went through Duffel’s direct airline connection, with loyalty numbers and cancellations supported. That airline has a defined commercial relationship with the agent.
When Skift asked the same agent for a hotel room in New York, it opened its own browser and picked between Expedia and Hotels.com. Nobody asked it to. The hotels were shopped by software they never signed a deal with.
That is the whole story in two searches. One side turned up early. The other waited, and the OTAs filled the gap. We have seen this before, when AI agents struggled with hotel websites and the OTAs picked up the booking. Waiting doesn’t keep you neutral. It hands your default position to whoever did show up.
Turning up doesn’t mean you win either
I don’t want to pretend that plugging in fixes everything. Skift points out that Google’s agentic hotel booking in AI Mode keeps the hotel or OTA as merchant of record, but passes on only the guest’s name, payment, occupancy and property. The conversation that led to the booking stays with Google.
So an integration buys you a place in the transaction, not a relationship with the guest. Skift’s advice is to price it as distribution access with little customer data attached. I agree, and it is exactly why I argued that Google wrote the rules for AI hotel booking while hotels didn’t get a seat. The terms you accept now are the terms you will be stuck with when volume arrives.
Payments and loyalty are being decided at checkout
The quieter fight is over who holds the wallet. Skift reports that Muse pays airlines with a one-time virtual card through Link by Stripe, and Google settles through Google Pay.
Delta collected $8.2 billion from American Express in 2025, around 14% of adjusted operating revenue. Co-brand cards are a loyalty engine. If agents default to their own wallet at checkout, they get a say in which card pays for the trip, and that money moves.
Loyalty itself is exposed too. Skift Research found 44% of US travellers book outside their preferred programme on price, and 55% would follow an AI tool to a better deal. Loyalty membership ranked last as a reason to trust an AI recommendation. A points discount is just a price, and agents are very good at comparing prices. That fits what I wrote last week: AI isn’t killing loyalty, interchangeable brands are.
The plumbing is already here
This isn’t only a game for Booking Holdings and Google. This week roommaster, a PMS provider for independent hotels, launched a connector that links live PMS data to ChatGPT, Claude, Copilot and Gemini using the Model Context Protocol, an open standard for plugging AI assistants into business systems. Right now it is for staff asking questions, not guests booking rooms. But it shows the pipes into the big assistants are being built at every level, not just at the top.
What I’d put in the 2027 budget
- Budget for position, not volume. Treat AI spend as buying a seat at the table, not as a channel that must pay back in six months.
- Know which agents can actually book you. Test the major assistants on your own rooms and fares, and fix what breaks.
- Decide your red lines on data. Know what guest data you need from an AI booking before you sign up to any programme.
- Stop selling loyalty as a discount. Recognition on property is something an algorithm can’t list or undercut.
- Talk to your payments and loyalty teams together. The checkout is now a loyalty decision.
The big players aren’t waiting for proof. Booking put $100m into an AI trip manager to buy position while it’s cheap.
So here is the debate. Is “wait and see” a prudent response to a 1% channel, or is it the travel industry’s most expensive way of saying “someone else can set our terms”?
If you want more takes like this on AI, search and travel marketing, subscribe to my LinkedIn newsletter.