A.I. Is Rewriting Travel Discovery. Can Booking Platforms Keep Up?

Observer ·

A.I. Is Rewriting Travel Discovery. Can Booking Platforms Keep Up?

A.I. can generate an itinerary in seconds. But can it understand what a traveler actually wants? Arie's Dmitry Buenkov argues that the industry's next A.I. opportunity is not simply automating reservations, but making recommendations that reflect individual taste, context and trust.

Last week, Sonder founder Francis Davidson announced that OpenAI had acquired Odessia , his A.I. travel agent, and then shut down the product. The move suggests that OpenAI was more interested in the team and its travel expertise than the app itself. It sees an opportunity earlier in the journey, when an assistant helps someone decide where to go. That is where the industry’s attention is shifting.

People already turn to ChatGPT or Claude for research, advice and routine tasks. Travel companies now have to compete with that habit. A.I. assistants are increasingly helping travelers figure out what kind of trip they want before they start comparing hotels or looking for flights.

Booking.com might still get the reservation. A traveler could choose a hotel with an assistant and pay through Booking.com. But if that traveler starts with the assistant again on their next trip, it gets the first chance to shape the shortlist. Booking.com may still process the transaction, but it has less influence over which properties the traveler considers, which alternatives they compare and what additional services they buy. The company that controls discovery has an opportunity to build the relationship over time, while the company that processes the booking may be left with a smaller role in it.

The major booking platforms are already responding. In July, Expedia acquired Layla AI, a trip-planning assistant, to combine it with its own data and booking services. Booking Holdings launched Vayo in private beta , giving the assistant access to travel confirmations in a user’s inbox, including bookings made through other sites. These platforms know what people searched for, compared, booked and canceled. But a booking alone doesn’t explain whether someone chose a hotel for its price, location or atmosphere. If travelers share those preferences with an assistant, it may learn things the booking platform never hears.

Google is pursuing another model. In August, it began rolling out hotel booking within A.I. Mode in the U.S., in partnership with Booking.com, Expedia and hotel chains. Travelers can select a hotel through Google, while the partner handles payment and customer service. That arrangement could generate additional bookings. But if it replaces direct visits to booking platforms’ own sites, those companies risk paying to reach customers who once came back directly. The economics will depend on the terms of each partnership. Either way, the assistant gains influence over which businesses get the booking and, potentially, which properties get seen in the first place.

For a booking platform, losing that influence has consequences beyond a single reservation. Discovery is where a platform can put higher-value properties in front of travelers, encourage them to consider alternatives and introduce additional services. If an assistant controls the shortlist, it may also control those opportunities. A booking platform could still collect the transaction, but have less say in how the customer gets there or what else they buy along the way.

Hotels themselves face a different challenge. A property can have rooms available and still miss the traveler’s shortlist. It now needs to explain to A.I. who it is best suited for. “Cozy hotel in the city center” says little about whether it better fits a couple seeking a quiet evening or a group of friends looking for a lively stay. Assistants need more specific information to distinguish one property from another.

Shaping the shortlist gives a service influence over the eventual booking. Earning that role depends on how well its recommendations fit the traveler.

A.I. can be generic by default. During a user interview, I watched a participant ask an early A.I. travel planner for ideas for a trip to London in November. He had specified interest in brutalist architecture but still received recommendations for Big Ben and Covent Garden. Asking again added the Barbican without removing the original suggestions. Then came flight recommendations, suggested before the assistant had asked about his transport preferences. He closed his laptop. One test isn’t a verdict on the product, but it illustrates why an assistant needs to know when to ask questions rather than generate more suggestions.

Ratings also miss what matters to the individual traveler. For someone who buys coffee beans, a café’s selection of beans to take home might matter more than its overall rating. Memory helps, but it isn’t enough. Mindtrip says its new assistant remembers preferences between trips, yet a work trip and a weekend with friends call for different recommendations.

Trust in the source matters, too. A tip from Monocle, the Michelin Guide or a friend with similar tastes can matter more than stars. And a bad pick can be costly. On Spotify, you skip a track in seconds. In an unfamiliar city, a poor recommendation can cost you an evening of a short vacation.

That sets the bar for any company competing in travel discovery today. A specialist app needs to make choices easier than the assistant people already use. Why pay for a separate service if you still have to check and redo its recommendations? For larger platforms, the challenge is similar. Having the inventory and processing the payment may not be enough if another service has earned the traveler’s trust before the booking begins.

Over the next year or two, I expect more products that let travelers ask questions, compare options on a map and keep track of their choices. Airbnb plans to begin rolling out its own agent , with multiple specialized tools for guests, in the first half of 2027. But a chat window beside a map isn’t enough. The system must distinguish what someone browsed, rejected and ultimately chose without asking them to explain their preferences all over again.

Payments are moving into the picture, too. On Nov. 13, OpenAI introduced ChatGPT Wallet , initially for U.S. Pro subscribers, with users approving each purchase. Approving a payment is different from trusting an assistant’s travel advice. That trust has to come from recommendations that consistently fit the person. If an assistant earns it, it may eventually make the booking itself. The payment may still go to Booking.com, but the customer relationship could belong to the assistant. That may be the bigger threat.

Travel companies need to rethink the customer journey: when and through which channels travelers discover them, and how to win back that first point of contact. To bring travelers back directly, market players need to figure out how to be useful long before those travelers have decided what to book.

Dmitry Buenkov is the co-founder of Arie , a Paris-based travel discovery app. He spent 15 years building digital products and design systems across technology and creative industries. Today, he builds Arie, an app that recommends places and things to do based on personal taste and the context of each trip.

Источник: Observer