By Tom Whitfield

AI-driven referrals still account for less than 2% of traffic at both Booking.com and Marriott, even as they grow exponentially. That figure, from Bernstein analyst Richard Clarke, anchored the Skift Data + AI Summit Europe, where executives agreed on one point: the agentic booking conversation has outrun operational readiness.

"The agents are here," said Seth Borko, Skift's Head of Research. "Almost no one [in travel] is ready for this world."

Agentic booking tools such as Instinct and Muse have entered the market and consumer adoption is accelerating. But the summit framed five operating decisions travel companies now face on AI investment, exploration, and execution — with agentic AI booking volumes still small.

What decides which hotels AI systems recommend?

Research from Stiplo founder Carlo Del Mistro quantified the discovery gap. Drawing on more than 27,000 travel queries across 650-plus Barcelona hotels, the study found independent hotels were eight times more likely than chains to be unknown to an AI system.

Once an AI system recognized an independent property, it recommended it at rates similar to chained competitors. The barrier is visibility, not quality.

Two findings carry direct revenue implications for sellers:

• Google review volume was the strongest driver of whether an AI system knew a hotel at all.
• llms.txt files — the much-discussed technical fix for AI visibility — showed no measurable advantage.

For independents, the takeaway is plain: review volume on Google, not AI-specific web plumbing, currently determines whether an agent can surface your rooms.

Why convenience won't carry agentic booking

Kayak CEO Peer Bueller and Skyscanner CEO Bryan Batista both argued that convenience alone will not convert travelers to agentic booking. The unresolved barriers are trust, selection, and price verification.

Skyscanner's research found 51% of surveyed travelers have concerns about the reliability of AI recommendations. Comparison complexity — the core of how metasearch sells today — remains hard for agents to reproduce credibly.

The number matters for distribution: if accuracy doubts persist, agentic channels cannot simply inherit the conversion rates of existing search and metasearch funnels.

How are operators governing AI spend?

Accor, IAG, Strawberry, and Amex GBT described a common pattern: narrowing AI projects to a small number with visible returns, and owning infrastructure only where control matters.

EU AI Act governance did not stop experiments, speakers said — it pushed them into a structured review process before production deployment.

Amex GBT offers the sharpest measure of that shift: the corporate travel giant reviewed roughly 139 AI tools or use cases in the past year, up from 13 the year before. That is a tenfold increase in formal evaluation volume at one of the largest travel management companies.

Bernstein's Clarke added a market-level caution: no clear AI winner has emerged in consumer industries, despite heavy investment across sectors.

The structural position — sub-2% referrals at Booking and Marriott, an eightfold visibility penalty for independents, and majority traveler skepticism — suggests the agentic shift is real but early, and sellers who treat visibility and trust as measurable variables will price the transition better than those waiting for the channel to arrive.