Search twice, book once

The first AI trip planners have arrived, but they still have a long way to go.

Everyone who has planned a trip knows that the process splits into two stretches that feel nothing alike. You spend the early one browsing destinations, imagining itineraries, and accumulating possibilities that may never survive contact with a calendar. The late one has you comparing fares across multiple tabs, verifying that hotels exist and have availability, and reading cancellation policies before handing over a credit card. Most of the frustration that travelers associate with planning lives here. Over the past year, Google, Expedia, and a tier of specialist startups have launched AI tools aimed at the process, and these tools have concentrated almost entirely on the half that was already pleasant.

Vacation dream machines

Google launched Flight Deals last August as a Gemini-powered addition to Google Flights. Travelers describe a trip in natural language instead of filling in the usual grid of dates, airports, and cabin class. A query like “week-long trip this winter to a city with great food, nonstop only” returns destinations that might never have occurred to the traveler, alongside real-time fares from more than 300 airline partners. The AI interprets the request, identifies matching cities, and surfaces pricing from across Google’s inventory. Google expanded the tool to more than 200 countries and 60 languages last fall.

Expedia launched Trip Matching last May, and the feature lives entirely inside Instagram. A traveler who spots a reel of glass igloos under the Northern Lights shares it to Expedia’s account via DM. The AI analyzes the video, identifies the location, and replies with a destination guide that includes suggested timing, an itinerary, and hotel recommendations with prices and ratings. Eighty percent of millennials rely on social media for travel decisions, according to research that Expedia cited when it announced the feature, and Trip Matching intercepts that inspiration inside the app where it originated.

These tools serve a phase of trip planning that most people already enjoy. Exploring where to go next has always felt closer to recreation than to labor. A TakeUp survey from earlier this year found that 90% of travelers know that AI can help plan travel, while only 38% have tried it. The gap suggests that most travelers have not yet felt an urgent need for help in this particular phase. AI has made the dreaming faster and more serendipitous, and in doing so has improved a process that was already pleasant.

Serviceable itineraries

Once a traveler settles on a destination, general-purpose chatbots become the default structuring tool. Head-to-head comparisons of ChatGPT, Gemini, Claude, and DeepSeek consistently find that all of them can produce a serviceable day-by-day itinerary from a decent prompt. ChatGPT tends to offer the most optionality, listing multiple restaurants and activities for each slot in the day. Gemini integrates tightly with Google Maps and Sheets, and one tester found that it tailored family-travel recommendations better than competitors, suggesting minivans instead of compact cars for a trip with two kids without being asked. DeepSeek tends toward practical specifics that others skip, like local grocery stores and tour-operator names.

A tier of specialist tools has filled in the remaining gaps. Layla AI, which acquired the itinerary platform Roam Around and has generated more than eight million trips to date, pulls live pricing from Skyscanner and Booking.com. Mindtrip adds map-based visual planning with integrated reviews. Kayak’s integration with ChatGPT brings real-time booking-database searches into the conversational interface. Gondola occupies an even narrower niche, calculating whether a hotel is cheaper as a cash booking or a loyalty-point redemption across more than 700,000 properties.

Amadeus coined the term “Travel Mixology” in its 2026 trend report for the emerging behavior of layering these tools together. A traveler might consult a chatbot for broad research, check Reddit or YouTube for firsthand accounts, then use a brand’s AI assistant for suggestions built on past search behavior. An IMG survey released last week found that 33% of travelers plan to use AI tools this year. Among that group, 75% intend to use them for recommendations, 70% for itineraries, and 69% for discovering new ideas. The adoption curve steepens with experience. TakeUp’s research found that among travelers who have actually tried AI for planning, 94% trust the recommendations as much as traditional sources.

Collision with reality

Multiple independent reviews of AI travel planners report consistent failures at the point of transaction. One tester found that Expedia’s Trip Matching defaulted every booking link to a generic New Jersey results page instead of the intended Lapland listings. AFAR’s review found that Gemini declined to provide booking links until asked twice, and that Kayak’s ChatGPT integration had to be told repeatedly to find hotels under a stated budget before producing results. Chatbots hallucinate restaurants that do not exist and suggest flight connections that cannot be made. These failures cluster at the transactional layer, where a broken link or a fabricated hotel carries real consequences.

The survey data quantifies this confidence gap. Among travelers who plan to use AI this year, only 13% would use it for actual booking, a fraction of the large majorities who would use it for recommendations and itinerary building. IMG’s chief commercial officer noted that travelers rely on AI during the exploration phase but revert to established brands and direct research when the moment of booking arrives. This confidence gap maps onto the architecture of the tools. General-purpose chatbots generate text fluently but lack persistent connections to inventory databases, real-time availability, and payment systems.

Google and Expedia are both working to close this boundary. Google announced late last year that it would enable flight and hotel booking directly within AI Mode, in partnership with Booking.com, Marriott, Wyndham, IHG, and Choice Hotels. Expedia’s Trip Matching already embeds hotel cards with prices and ratings inside Instagram DMs. In both cases, clicking through to book opens a browser window and a traditional interface. Travel industry analysts noted that OTA stocks dipped when Google’s plans were announced, a signal that the industry expects this boundary to erode. Travelers still compare prices across multiple tabs, verify that hotels and restaurants exist, and read cancellation policies by hand. The phases of planning that lead up to booking have been transformed, and the booking itself has barely changed.

Not there yet

This asymmetry extends well beyond travel. AI in its current form excels at generating prose, synthesizing options, and recombining known information into new arrangements. It struggles at tasks that require verifying facts against the physical world in real time. A chatbot can describe a dozen ways to spend a Tuesday in Lisbon. Confirming that a particular restaurant is open on Tuesdays, accepts reservations, and has a working booking link requires a different kind of reliability. Language models do not yet provide that reliability consistently.

Travelers have responded accordingly. The “Travel Mixology” pattern that Amadeus identified describes people who use AI as one layer in a multi-source workflow, consulting chatbots for structure and scope, then switching to Reddit for the texture of firsthand experience, then opening a booking site to verify prices and complete a transaction. This hybrid behavior treats AI as a collaborator in a larger workflow, and it may prove more durable than the seamless end-to-end automation that Google and Expedia are racing to build. The most useful travel tool in 2026 is a chatbot and six open tabs.