Amex GBT reviewed roughly 139 AI tools or use cases in the past year, up from 13. That tenfold jump framed the operating reality at the Skift Data + AI Summit Europe, where travel executives confronted a shift they cannot slow down and largely have not budgeted for.

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

Agentic booking (AI agents that search, compare and complete transactions on a traveler's behalf) remains small in volume. But the AI labs building it are moving faster than the travel industry's governance, pricing and distribution plumbing. The summit organized the response around five operating decisions covering AI investment, exploration and execution.

Who gets seen by the machines?

The hardest data point came from Carlo Del Mistro, founder of Stiplo. His study ran more than 27,000 travel queries across 650-plus Barcelona hotels and found that independent hotels were eight times more likely than chains to be unknown to an AI system. Once an AI model did know an independent property, it recommended it at similar rates to chains.

The visibility gap, in other words, is a discovery problem, not a quality judgment. Del Mistro's findings point operators toward concrete levers: Google review volume was the strongest driver of AI visibility, while llms.txt files (a proposed standard for telling AI crawlers what to read) showed no measurable advantage. For independents, the cheapest AI strategy may simply be review generation. The technical workaround, at least in Barcelona, did nothing.

What actually blocks agentic booking?

Executives from the metasearch side rejected the framing that convenience will drive agent-led transactions. Kayak CEO Peer Bueller and Skyscanner CEO Bryan Batista argued that trust, selection and price verification remain unresolved. Batista cited Skyscanner research showing 51% of surveyed travelers have concerns about the reliability of AI recommendations.

The economics cut both ways. Agentic booking promises lower acquisition costs for suppliers but compresses the comparison and merchandising surface where brands currently differentiate. If agents verify price rather than browse shelves, operators lose the display real estate they pay for today.

How do operators govern the spend?

The most instructive disclosures came from large operators. Accor, IAG (the airline group behind British Airways and Iberia), Strawberry and Amex GBT all described narrowing their AI portfolios to a handful of projects with visible returns, after an initial phase of broad experimentation. Owning infrastructure, they said, makes sense only where control genuinely matters.

Regulation shaped the process rather than stopping it. The EU AI Act, the bloc's risk-tiered AI law, pushed experiments into a formal review process before production deployment — a governance tax that Amex GBT's 139-tool figure makes concrete.

Has anyone won yet?

Bernstein analyst Richard Clarke offered the corrective to summit enthusiasm. No clear AI winner has emerged in consumer industries, he said, and AI referrals still sit below 2% at both Booking Holdings and Marriott International, even as the numbers grow exponentially. Exponential growth from a small base is still a small number.

That is the operator's dilemma in one line. The share of bookings and referrals arriving via AI is too small to reorganize a distribution budget around, yet the growth curve is too steep to ignore.

What happens next?

The five decisions — where to invest, what to build versus buy, how to govern, how to stay visible, and how to price for agent-mediated transactions — will not resolve this year. But the Barcelona data suggests visibility gaps close through reviews and reputation, not technical file standards, and Amex GBT's review volume suggests governance, not capability, is now the binding constraint on deployment.