Winning the Decision Layer: What AI Changes for Hotel Distribution

This 37-minute session from ITB Berlin 2026 opens with an impulse keynote by Sanjay Vakil (CEO, Direct Booker) followed by a panel with Uta von Dietze (VP Commercial Performance EMEA, Wyndham Hotels & Resorts) and Andrew Boch (CEO, Moi AI), moderated by Lea Jordan. The core argument: AI is not a future threat to hotel distribution — it is already restructuring how guests discover, decide, and book, and hotels that fail to act now risk ceding the top of the funnel to OTAs for another 20 years.

Vakil anchors the session in concrete data. Citing SemRush research, he notes that queries submitted to Google's AI tools are three times longer than traditional keyword searches — travelers are typing nuanced, full-sentence questions rather than compressed keywords like 'hotels Berlin.' Google has also recorded a double-digit percentage increase in searches for individual hotel names, which Vakil hypothesizes reflects a new behavior: users research on AI, then pivot to a direct hotel search. A Phocuswright dataset he references shows explosive growth in generative AI usage specifically for travel, with the growth curve described as 'astonishing.'

Von Dietze adds operator-side data: 40% of consumers now use large language models (LLMs) for travel search, and 95% of those users say they will use an LLM again. Crucially, the share of LLM researchers who actually complete a booking through that channel rose from approximately 1% a year ago to 12% — an 11-percentage-point increase in roughly 12 months. This signals a rapid maturation from inspiration-only to transactional use of AI for travel.

Vakil announces live during the session that Direct Booker is now available as a connector on Claude, enabling end-to-end hotel booking — search, selection, and direct completion — without the user ever leaving the AI interface. He frames this as the beginning of closing the loop between AI-driven discovery and commission-free direct bookings, at a price point he claims is roughly 10x cheaper than OTAs.

Boch identifies the next 6–12 months as a critical window for commerce enablement inside LLMs. Until recently, AI platforms kept users in the 'inspiration phase'; now transactional flows are opening. He recommends hotels immediately deploy an MCP (Model Context Protocol) server and an MCP app or ChatGPT plugin. He describes MCP as a method for connecting first-party hotel data directly to LLMs, creating a branded interface inside AI surfaces. His view: any hotel with a website can and should have an MCP app live within three months.

A recurring theme is the infrastructure gap. Vakil notes that GDS note fields are as short as 20 characters — structurally incapable of carrying the rich contextual data (e.g., hypoallergenic soap preferences, balcony requirements, pet-friendliness, interconnecting rooms) that AI conversations now elicit. Boch adds that existing analytics stacks are built around Google Search click-conversion funnels and cannot capture the semantic richness of AI interactions. His firm is building parallel analytics infrastructure for clients as a stopgap.

On the competitive question of who benefits most, Vakil argues large chains like Wyndham (8,300+ hotels) can deploy improvements instantly across all properties once built. But Boch counters that smaller independent hotels, with shorter chains of command, may actually produce richer, faster property-level data — a potential advantage in AI aggregation, where data richness beats checkbox volume. Von Dietze confirms Wyndham has spent several years migrating to cloud infrastructure to enable exactly this kind of AI-readiness.

The panel's closing warning is pointed: LLMs learn from observed consumer behavior. If OTAs continue to be the surface where bookings happen, the models will recommend OTAs. Hotels must appear — with better prices and richer offers — on AI surfaces now to shape those training patterns before they calcify into the next generation of consumer defaults.