航空业准备好接受代理主导的预订了吗？

To stay competitive, airlines need to appealto AI—not just human users. By Rostislav Khomenko and Geoffrey Weston Is the Airline Industry R eady for Agent-Led Bookings? At a Glance Agentic artificial intelligence (AI) is reshaping the front end of travel, assisting travelers withflight discovery and research. Soon, agentic tools could help travelers book flights. We conducted two tests to determine whether online travel agencies (OTAs) and airline carriers In our tests, OTAs showed clear advantages in both agent-led discovery and booking. To remain competitive, airlines need to adapt their commercial and technical foundations toappeal to large language model agents—not just human audiences. Agentic artificial intelligence (AI) is reshaping the front end of travel. Unassisted, travelers often struggle with choice overload, compare too few dimensions, or fixate onheadline price. AI agents, by contrast, can normalize trade-offs across fares, ancillaries, and constraints These differences fundamentally change how travel options are curated and selected. When discoveryand decisions are delegated to machines, transparent, dynamically bundled offers are likely to outperform The evolution seems inevitable. In the next wave of agentic AI, booking could take place with minimal Is the industry ready for this agentic travel future? To find out, we ran two practical tests: one focused onflight discovery and one on flight booking. Flight discovery test results For the flight discovery test, we selected three large European full-service carriers (FSCs) and their 10most relevant markets (aggregated at the country-pair level). Then we prompted three major largelanguage models (LLMs) to book a flight from country or city A to country or city B, using 60 This test measured two things: which sources LLMs used to search schedules and prices; and which In this test, airline websites were accessed directly only about 5% of the time(see Figure 1). Is the Airline Industry R eady for Agent-Led Bookings? Figure 1:When prompted to find flights, LLMs directed travelers to OTAs farmore frequently than airlines Subsequently, we reviewed how often airlines appeared (on their 10 major passenger flows) in answersproduced by the models. All three European FSCs consistently showed up in the top three results whenqueries involved routes touching their home markets. However, their share of mentions declined Moreover, LLMs frequently directed users to an OTA rather than an FSC website—even when an FSC was The implication is clear: LLMs gravitate toward the source with the easiest downstream interaction, andOTAs currently produce cleaner, more structured, and more agent-readable data. For airlines, this creates a generative engine optimization challenge. Airlines need to rethink how offersare surfaced, structured, and interpreted by LLMs rather than humans. Today, OTAs are clearly winning Is the Airline Industry R eady for Agent-Led Bookings? Figure 2:LLMs are more likely to recommend airlines when travel is to or from Prompt: “I want to book a light from [city 1] to [city 2].(Please mention one or several airlines in your answers.)” Distribution of rankings in search results produced by LLMs Booking test results Next, we attempted to book travel with several major European and North American airlines, includingFSCs and low-cost carriers (LCCs), as well as two major OTAs. We used three tools: OTA action tools/plug-ins, browser automation agents, and prompt-generated airline links. A booking was considered successful if a tool reached the payment page of an airline or OTA website.Across our attempts, no single tool was able to do so reliably—although OTAs were more successful(see Figure 3). Is the Airline Industry R eady for Agent-Led Bookings? Some browser agents reached deep into the flow, but only slowly and erratically, and user interventionwas frequently required. Browser agents regularly stalled on basic user interface (UI) components such asdate pickers or fare selectors, looping for 10 to 20 minutes before abandoning the attempt and redirecting These results are unsurprising. These websites were designed for humans, with cybersecurity measures To compete in this space, airlines need to quickly adapt their commercial and technical foundations.Otherwise, they risk structural disadvantages as agentic infrastructure matures. For example: •OTA action tools/plug-ins:These function primarily as search widgets that return lists of flights but •Browser automation agents:Autonomous browsing agents could navigate airline and OTAwebsites but were slow, inconsistent, and error prone—often taking 5 to 30 minutes before getting •Prompt generated airline links:LLMs often claimed they could generate links with traveler detailsprefilled, but consistently failed to do so. At best, they produced static search links. Is the Airline Industry R eady for Agent-Led Bookings? Where do airlines go from here? AI agents favor su