AI-driven search and booking is a new redistribution wave, after the GDSs, the OTAs and mobile. This page collects the external evidence behind that proposition: publications from strategy firms, analysts, institutions and market-data vendors, grouped by board question and ordered from load-bearing evidence to background. Each entry gives our short take, one sourced number and a link to the original publication.
Where will the moment of choice move, and who captures the booking?
The only named European operator in this library putting a number on its own AI exposure. Sunweb's Head of Marketing Channels reports orientation journeys collapsing from ten to fifteen clicks down to four or five. Crucially, total site traffic still grows: this is redistribution of the moment of choice, not demand disappearing.
20% year-on-year decline in organic search traffic; orientation journeys down from 10-15 clicks to 4-5.
Core source
Skift's fifth annual industry benchmark, rebuilt around a four-layer travel stack. Its sharpest finding for suppliers is a visibility gap: most hotels simply do not appear at the first traveller query. Discovery has moved from search-scroll-compare to ask-shortlist-decide.
Only 6% of hotels appear in AI-generated search results; traveller familiarity with AI trip planning at 62%, usage up 124% year over year.
Core source
Frames agentic AI as the next redistribution wave after GDS, OTAs and mobile — the structural backdrop for the whole assessment.
AI mentions in travel annual reports rose 4%→35% (2022–2024); >90% trust AI travel info, but only 2% let AI book.
Empirical LLM tests: agents route discovery and booking to OTAs, not suppliers — hard proof that AI redistributes demand to intermediaries unless offers are machine-readable.
Airline websites accessed directly only ~5% of the time in agent-led discovery.
Freshest, most on-thesis channel numbers: hotels barely surface in AI search, and OTAs/editorial dominate AI recommendations.
Only 6% of hotels appear in AI search results; 82% of AI hotel recommendations originate from OTAs and editorial media.
Travel-native analyst framing agentic AI explicitly as a distribution question — authority citation alongside the harder data.
"Agentic AI will redefine the travel marketplace as profoundly as the internet once did."
What does AI do to commissions, distribution costs and direct bookings?
Core source
The clearest published match to the proposition: OTA commissions, emerging AI-era distribution fees, machine-readable content, and "own the guest relationship — and the margin."
OTA commissions 15–30%; digital direct bookings ($262bn) nearly matched OTAs ($266bn) in 2024; 37% of travelers already use embedded LLMs to book.
The moment the agentic booking question stopped being hypothetical. Google confirmed a limited US test inside AI Mode on 7 August 2026, with Amadeus as the first GDS wired into its Universal Commerce Protocol. The structural point for independents: whichever pipe carries the booking decides whose guest it is when the agent is done.
Live test partners include Booking, Expedia, Marriott, Wyndham and IHG; Amadeus first GDS on Google's Universal Commerce Protocol, Sabre on the air side.
Google's stated position as of November 2025, kept as the record of what was promised. It is now overtaken by events: by August 2026 Google was completing hotel bookings in conversation. Read it against the agentic booking test above to see how fast the stated boundary moved.
On the announcement, Booking Holdings fell 4% and Expedia 7%; supplier site traffic expected to decline into Google's ecosystem.
Operator voice: Accor's CDO expects gen AI to reduce hotel intermediation and enable per-guest pricing. Use the one distribution quote; the rest is design/sustainability.
"In the future, some of this intermediation might disappear… a generative-AI-powered marketplace… customer A and customer B won't see the same offers."
The longest-running independent measurement of the previous redistribution wave: a biennial survey of 3,000+ European hotels tracking channel shares since 2013. The baseline against which the AI shift will be measured.
OTA share of European hotel bookings rose 19.7% (2013) to 29.6% (2023); direct fell 57.6% to 50.9%; Booking Holding holds ~71% of the OTA market.
Shows AI booking as live production infrastructure: a major channel manager pipes live rates into ChatGPT and Claude via MCP. (Vendor announcement.)
8 in 10 travelers want AI assistance during booking (Changing Traveller Report 2026).
Who will own the relationship with, and the data of, your guest?
Names the asymmetry precisely: the agent reads urgency, location and browsing behaviour, the channel running the agent keeps that signal and the price premium, and the hotel receives a clean, contextless reservation. Makes direct the only channel where the guest cannot be priced off their own data.
Hotels receive their listed rate while the channel pockets the agent-driven premium; the hotel is the one party to the sale flying blind.
Best source for the 'who owns the guest' pillar: banks and platforms compete for the loyalty relationship as members grow disloyal. (AI is secondary here.)
Median share-of-wallet for a traveler's 'preferred' brand is only ~50% (lodging) to ~60% (airlines/cruise).
Sharp board takeaway: agentic AI adds a layer rather than removing one — OTAs become inventory suppliers while the platform owns the guest and the journey data.
Google's initial agentic partners include Booking.com and Expedia — the two largest OTAs — contradicting the disintermediation narrative.
The GEO-vs-SEO layer travel + code positions one level above: agents disintermediate brands from customers. Useful to show what the assessment sits on top of. (Cross-industry.)
Up to 45% of shoppers already use GenAI to compare products; Alphabet reports 75M daily 'AI Mode' users.
How fast is this really moving, and what do travellers trust AI to do?
Core source Counterweight / nuance
Read this next to the Sunweb figure: together they separate the two curves. The traffic shock is real and already measurable, while the booking shift is not. Booking Holdings acknowledges AI Overviews pressure on SEO in the same quarter that agent-driven bookings stay negligible. Plan for both curves, but do not confuse them.
LLM referrals under 1% of room nights in Q2 2026, with no significant change in that share; room nights up 5% year over year.
The consumer-side counterpart to Phocuswright's business-adoption survey, and the first source here that describes who the AI traveller is commercially. They are not a fringe segment: they earn more, travel more often and spend more per year than non-users. Losing visibility with them is not a proportional loss.
56% used AI for at least one trip (from 33% in H1 2025); AI users' median household income $129,200 vs $104,000, and $4,500 annual leisure spend vs $3,000.
The adoption-speed datapoint for the stat-strip — how fast the industry is moving on agentic AI. (Adoption, not channel impact.)
61% experimenting or scaling (6% actively scaling, 22% beginning); 50%+ exploring MCP/A2A interoperability standards.
Counterweight / nuance
The honest counterweight: full-autonomy booking adoption is tiny. This nuances the thesis and makes the scenario story credible — measure and prepare, don't chase hype.
Only 2% of leisure travelers are willing to let AI book on their behalf.
A dense, cited stat sheet on the awareness-vs-use gap and model limits; thesis is that structured, machine-readable data decides who agents can even see. (Vendor aggregator — cite primaries.)
90% of US leisure travelers aware AI can book travel, but only 38% have used it; models score ~10% on complex multi-stop itineraries.
Quantifies the pilot-to-production gap — the practical starting point for an assessment — and names channel management as an AI impact area. (Thin, Middle-East focused.)
91% piloting or using AI, but only 3% at enterprise-wide deployment.
Counterweight / nuance
A Dutch-market counterweight with concrete failure modes rather than abstract trust scores. Adoption is broad but the output is not yet reliable, and the gap is worst on recency: five of six tested models did not know Bulgaria adopted the euro on 1 January 2026. Useful for sizing how much of the trust gap is fixable content quality.
52% of Dutch travellers used AI to plan or book; 46% of those received incorrect information or faulty tips.
Which numbers in your own funnel already show the shift?
The empirical backbone for 'measure your own numbers': real hotel-website analytics on AI as an emerging, commission-free, higher-converting channel. This is literally week 1–3 of the assessment.
AI referral traffic <1% of hotel visits but +50% in weeks; AI's share of organic grew ~2%→3.1%; AI referrals bypass 15–20% OTA commission.
Core to the scenario: the threat is discovery, not the transaction — the moment of choice moves upstream, before the booking site. Exactly where to measure.
Travelers using AI 'extensively' for trip planning rose 13%→30% (+124% YoY).
Vendor commentary on a June 2026 Aven Hospitality and h2c study, included because it is the only source here that turns agent-readiness into operational thresholds rather than forecasts. Its measurement point is the one that matters: an agent failure shows up as a competitor's booking, not as an error in your logs, so human-era dashboards cannot see it.
11% can complete a booking and price inventory in real time; the readiness bar is p95 latency under 800ms plus room-level verified mapping.
Big-Four authority plus a near-paraphrase of the pitch: as gen AI grows, travelers use fewer other research sources, and continuous evaluation of how travelers search, decide and transact becomes a standing priority.
~25% of travelers used gen AI for trip planning in late 2025 — triple 2022.
An external checklist for the '24-month investment' board question: five moves (own agent, interface, trust/data, loyalty, hyperpersonalization). (Advice, not evidence.)
"We don't think one all-purpose AI agent will dominate bookings" — most companies should build their own.
Broader context; supports "AI is changing travel", less the channel thesis.
Foundational scene-setter on gen AI across discovery, booking and in-trip. Broad and pre-agentic; context rather than channel-shift evidence.
McKinsey estimates gen AI could unlock $2–4 trillion in annual value across sectors.
Operations- and frontline-focused discussion. Note: partly argues the best AI stays 'opaque to customers via traditional channels' — useful nuance, not channel-shift support.
Post-COVID, the travel sector employs 12% fewer staff; roles take ~5–6 weeks longer to fill.
Top-tier institutional macro frame naming AI itineraries, personalization and price prediction as sector-shaping. Authority and context; light on channel/margin specifics.
Global travel spending projected at ~7% average annual growth (2024–2034).