Booking.com keeps about one in every seven dollars you spend on a hotel — what exactly are you paying for?

This explores what booking aggregators actually deliver for their commission, and whether AI agents are wearing that justification down. One caveat first: the corpus doesn't break down Booking.com's 14-15 percent figure. The closest benchmark is Jaipuria's model, which treats about 12 percent as the 'aggregator take' an AI agent might eventually reach in suitable categories How could an AI agent's revenue per transaction grow over time?. What the corpus does have is a clear picture of which parts of the aggregator role are holding up and which are eroding.

The main argument is that search was never the part being paid for. Prieto notes that Expedia, Booking, and Airbnb all posted strong stock gains in 2026, despite years of warnings that AI would cut them out. Their explanation is that AI makes discovery cheap but leaves payment, hotel supply, trust, and completing the transaction where they were, and OTAs (online travel agencies) control all of those Are AI agents actually disintermediating online travel platforms?. Goswami's data supports this. Independent hotels' reliance on OTAs rose to 63.4 percent even as agents appeared, because OTAs stay the 'merchant of record': the party that legally takes the money, handles refunds, and absorbs the risk. Expedia's business-to-business segment, which sells this plumbing to other companies, grew 24 percent and now makes up 38 percent of revenue Are AI agents actually removing online travel agencies from bookings?. So in this view, the take rate pays for settlement and liability, not for a search box.

The part that is eroding is the advertising-and-ranking business. Jarc argues that when an agent ranks hotels by the traveler's stated criteria, paid placement stops mattering. Tripadvisor's hotel revenue fell 23 percent in one quarter, with management blaming structural shifts in search Are AI agents making attention-based ads irrelevant for aggregators?. Trust data points the same way. Travelers are happy to let AI brainstorm trips but much less comfortable relying on it for high-stakes details like visas Do travelers trust AI differently for brainstorming versus visa tasks?. That gap leaves room for a trusted intermediary at the moment of commitment.

There's a less comfortable side to this. Take rates hide what's inside them. Jaipuria warns that merchant payments for placement can be folded into the commission, so users never see them. That gives an agent an undisclosed reason to steer recommendations Could agent advertising hide inside platform take rates?. Azhar makes a related point: an agent paid by Expedia's referral fees can't be fully loyal to both Expedia and the traveler Can an AI agent serve both merchant and user interests fairly?. A formal model suggests platforms can set fees in ways that move value from sellers to the platform without making buyers any better off Does platform pricing of inspection tokens encourage excess search?.

What you might not expect is that the leverage may be moving below the aggregators, not just to the agents above them. As agent searches approach 200,000 per ticket sold, Amadeus has stopped pricing every query live and serves cached, precomputed fares instead. That means whoever controls the cache and how often it refreshes partly decides what offers agents can see at all Does caching fares give infrastructure providers pricing power?. The take rate may hold up, but the fight over who earns it is spreading across more layers of the stack.