AI travel agents squeeze OTA margins, hotels race to adapt

Before AI began rerouting hotel demand, the OTA model had already been hardened into a familiar economic machine. “Between roughly 2011 and 2020, the rapid expansion of merchant and agency hotel distribution established the traditional OTA commission and booking flow models…,” with Expedia and Hotels.com associated with merchant economics and Booking.com with agency-style flows; that period, as the background notes, “directly preceded AI disruption,” making September 2026 significant not as a sudden invention but as the moment pressure started hitting a mature commission structure built for web-page traffic and intermediated checkout.

By then, traveler behavior and hotel strategy were already shifting in ways that threatened OTA margins. Hotelrevenueinsights News highlighted in early September 2026 that AI-led hotel discovery was extending into transaction fulfillment, while earlier analysis cited Skyscanner data showing “el 38% de los viajeros ya utilizaría IA para investigar un destino y un 33% para planificar completamente sus vacaciones,” evidence that AI-led planning was already mainstreaming; in the same September 5 guidance, hotels were explicitly urged to pursue direct demand by “moving business away from channels carrying 15–25% commissions,” a clear response to compressed intermediary economics.

September looked like an inflection because AI travel interfaces were no longer just another referral source; they were becoming the place where the trip itself gets assembled. Platform Aeronaut argued that discovery, planning, comparison and booking can now happen inside one conversational workflow rather than the old chain of search, metasearch and OTA tabs, and that shift matters because, in its words, booking becomes a routing decision rather than a browsing decision when the “router of intent” sits upstream in chat.

That centralization strips friction out of the consumer journey while also stripping leverage out of the OTA model, because the user can get an answer without ever landing on an OTA page. Good Morning Hospitality captured the market’s anxiety bluntly: the “new fear” is “that this next wave of AI agents is just gonna start booking direct, and that is going to… disintermediate the OTAs,” a threat reflected in moves including Booking.com down roughly 15% and Expedia down roughly 13%; separately, one analysis warned that if one in six OTA bookings goes direct over five years, OTA roomnight growth falls from 5% consensus to roughly flat.

Trip planning is moving inside conversational interfaces that return answers, not lists: Skift quoted Aven Hospitality’s Mark Hollyhead saying, “We’re moving from a world of search results to one of answers,” where visibility depends on being “the most relevant response.” That shift is already commercially meaningful, with 44% of travelers using AI-powered search treating it as their primary source of travel insight, while The Hotel Conversation said “AI chatbots have become the new front door to trip planning,” exemplified by a natural-language family trip query that yields an instant hotel shortlist.

Inside that new front door, hotels are being ranked by what AI systems can verify, understand, and act on rather than by paid placement or generic web rank. Hospitality Net reported that visibility is “concentrating fast, among a small group of winners,” with 5W’s AI Visibility Index showing “the top three brands taking the lion’s share in several categories,” even though “loyalty-program size did not predict who got cited,” while Hotelrank found “roughly 65% of hotel queries trigger a live web search,” meaning recommendations are assembled in real time from live sources and favor properties with executable booking capability—still rare, since Skift said only 11% of hotel organizations have deployed true AI agents capable of completing bookings.

Independent hotels can win direct-booking visibility in AI channels if they give conversational systems something reliable to read and something executable to do. Let’s Data Science says structured property data now determines whether AI can surface accurate options and push users into a hotel’s own booking flow, while API, MCP and PMS connectivity determines whether assistants can show live availability; without that plumbing, discovery often leaks back to OTA funnels instead of the property’s direct channel.

Visibility also depends on trust: consistent listings, complete Google Business Profiles, current amenities and room data, recent reviews, and verified identity signals reduce hallucination risk and make a hotel legible to recommendation engines. That matters commercially because AI-referred visitors recorded a 14.2% conversion rate versus 2.8% for conventional organic search traffic and were 4.4 times more valuable, while one speaker arguing AI will favor direct bookings long term said the inventory owner—the hotel—is positioned to be the “king and queen.”

The strongest evidence that this shift matters now, even before it dominates bookings, is that hotels are already budgeting for it as a commercial priority rather than a lab experiment. PR Newswire’s March headline — “Canary Technologies: 82% of Hoteliers to Boost AI Spend in 2026, Eyeing Revenue and Guest Gains” — shows adoption scaling beyond pilots, and the timing matters: “Boost AI Spend in 2026” suggests near-term pressure on distribution economics, while “Revenue and Guest Gains” implies owners expect measurable upside from AI-shaped demand and visibility, not just operational efficiency.

At the same time, the incumbents’ own numbers show why the debate is about timing, not direction. Booking.com’s CEO said AI agent use was at “penny adoption” but had “doubled every month for the past few months” and improved conversion, cancellation, and customer success, even if “it’s not really doing much” in totals yet against a base where “last year we did… 86 billion dollars worth of travel… over a billion room nights”; yet Skift reported Curacity and Cornell research finding “94% of hotels are invisible in AI search results,” a concentration shock that can pressure large, “big lumbering like machines” with heavy software costs.