Your Direct Booking Strategy Has a New Problem. AI Sends Travelers to OTAs First

Close-up of a person using a laptop on a wooden table to book a hotel room on an OTA website showing search results.

by

The Hotel Blueprint

The hotels that expected AI to liberate them from Booking.com and Expedia are watching those platforms post their best financial results in years. The commission problem did not change — the plumbing that routes travelers to it just got faster.

1. OTAs Are Winning the AI Transition — and the Financials Confirm It

The inventory moat AI could not cross

When large language models like ChatGPT, Gemini, and Perplexity began reshaping how travelers search for trips, the working assumption across much of the hospitality industry was that AI would compress or eliminate the traditional booking funnel — reducing the role of aggregators and routing travelers more directly to hotel websites. That assumption has not played out in the data.

Booking Holdings reported full-year 2025 revenue of $26.9 billion, a 13.4% increase year-on-year, with gross bookings up 12% and room nights booked up 8%, according to its 2025 Form 10-K filed with the SEC. GAAP operating income reached $8.83 billion, representing an operating margin of 32.8% on a GAAP basis, with adjusted EBITDA margin expanding to 36.9%, up from 35.0% in 2024. Expedia Group reported full-year 2025 revenue of $14.73 billion, up 8%, with Q4 2025 revenue of $3.55 billion growing 11% year-on-year, according to its Q4 2025 earnings release. Both companies accelerated room night growth through the year.

The structural reason is straightforward. AI platforms need to fulfil the travel recommendations they generate. To do so, they require structured inventory data — machine-readable rate and availability information that can be retrieved instantly and returned as a coherent answer. Booking Holdings maintains availability data on more than 4 million properties globally. Expedia Group has a comparable global lodging inventory supplemented by recent acquisitions in activities (Tiqets) and ground transport (CarTrawler). Aggregating that inventory has taken both companies decades and billions of dollars. An analyst at Jefferies, cited in a December 2025 report, noted that aggregating long-tail inventory remains the core unsolved challenge for AI platforms — a problem that Booking and Expedia have spent twenty years solving. When an AI assistant recommends a hotel in Lisbon or Kyoto, the easiest fulfilment path — in most cases, the only practical fulfilment path at scale — runs through one of these two platforms.

A new discovery layer, the same commission line

The commercial consequence is that AI has added a new discovery layer to the funnel without removing the commission layer beneath it. A traveler who begins their trip planning with a conversation on ChatGPT or Perplexity and ends up booking through Booking.com has still generated a 15–18% commission obligation for the hotel, according to commission rate data from Cloudbeds and industry distribution analyses. The discovery experience changed; the cost structure did not.

For independent hotels, the situation deteriorated in 2025. The Cloudbeds 2026 State of Independent Hotels Report — compiled from 90 million bookings across tens of thousands of properties in 180 countries — found that OTA share of independent hotel bookings rose to 63.4% in 2025, with some markets approaching 80%. That is the distribution context into which AI discovery now feeds.

Where the AI-era profits are going

The more significant financial signal is what OTAs are doing with their AI-era profits. Booking Holdings used its 2025 Transformation Program to realise $250 million in operational savings — primarily through AI-assisted customer service reductions — and is reinvesting those savings into marketing, platform speed, and inventory expansion rather than returning them to hotel partners through lower commission rates. Expedia’s CEO Ariane Gorin confirmed on the Q4 2025 earnings call that the company is working with “all the major platforms to capture traveller demand, ensuring our brands show up prominently in AI searches and function effectively with agentic browsers.” Both platforms are spending to become the default fulfilment layer for the AI era before the architecture calcifies.

2. The Cancellation Asymmetry That Compounds the Commission Problem

The cost that commission percentages do not show

Commission rates are the most visible cost of OTA distribution, but they are not the only one. The Cloudbeds 2026 report surfaces a metric that deserves more attention from revenue managers: OTA cancellation rates hit 21.8% in 2025, more than double the 10.6% rate recorded for direct bookings from the same property base. The gap has widened over successive years of the report.

This divergence is partly structural. OTA platforms have consistently competed on flexibility as a consumer proposition — free cancellation has been central to the Booking.com value offer for years, and the platform’s AI-enhanced ranking algorithms reward properties that offer it. Travelers booking through an OTA face lower friction to cancel than travelers who have paid directly, often with a deposit or pre-payment. AI-driven discovery may reinforce this pattern: a traveler who arrives at a booking through a conversational AI interface, which surfaced multiple comparable options, retains the psychological option to reverse the choice more easily than one who has spent thirty minutes on a hotel’s own website before committing.

RevPAR lost twice: once to commission, once to cancellation

The 11-percentage-point gap between OTA and direct cancellation rates has a direct RevPAR effect. A room that cancels 39 days in advance — the average cancellation window recorded in the Cloudbeds report, up from 35 days in 2023 — can in principle be resold. But reselling inventory that was committed to and released by an OTA booking involves operational cost: re-entry into the channel at potentially lower rates as the arrival date approaches, risk of displacement of higher-margin direct inventory, and in some cases, loss of the booking entirely if demand has softened.

The combined cost of OTA distribution — gross commission plus the margin erosion from higher cancellation rates, plus the absence of first-party guest data — creates a materially wider gap between OTA revenue and direct revenue than the headline commission percentage implies. This is not a new observation, but AI discovery is adding volume to the OTA channel while leaving its structural disadvantages intact.

The booking window is lengthening — that is the usable signal

Some operators have begun structuring their direct booking offer explicitly around the cancellation differential — offering flexible or equivalent cancellation terms via their own channel while preserving the ADR premium that direct bookings typically command. The Cloudbeds data suggests the average booking window is now 40 days, up from 38 in 2023, with North America and EMEA leading at 48 and 47 days respectively. A longer booking window creates more opportunity to intervene in the channel mix before arrival — but only for operators who are monitoring the signal.

3. AI Discovery Is Fragmenting — and OTAs Are Building Into Every Fragment

Four platforms, one problem: OTAs are integrating into all of them

The AI search market is not a single platform. As of April 2026, ChatGPT held approximately 63% of B2B AI referral traffic across major platforms, down from 89% eight months earlier, according to an analysis by Goodie tracking referral data from 41 brand sites. Claude reached 18.5%, Gemini 10.6%, and Perplexity 7.3% over the same period. The implication is that AI-driven travel discovery is no longer a ChatGPT question — it is a multi-surface distribution problem, with each platform carrying different retrieval logic, citation behaviour, and fulfilment architecture.

Expedia and Booking Holdings have responded by pursuing integration partnerships across all major AI surfaces rather than betting on any single platform. Gorin’s Q4 earnings call language — working with “all the major platforms” — is an operational statement, not a branding one: both companies are building API integrations and data connections into each of the major AI interfaces so that their inventory becomes the recommended fulfilment layer regardless of where the traveller begins their search. Booking.com’s October 2025 launch of agentic AI features for partner communication, and its integration into Google’s AI travel tools, represent the same strategy executed from the supply side.

Total AI referral traffic across all industries currently represents approximately 1% of overall web traffic, according to Conductor data reported by Digiday in December 2025. In travel, the proportion is higher and growing faster — but the absolute booking volume routed through AI-native interfaces remains a fraction of total transactions. The 3–5% AI booking share figure in major markets cited in some distribution analyses cannot be traced to a named primary source and should be treated with caution.

Invisible to AI, invisible to the guest

The fragmentation of AI discovery creates a compounding visibility problem for hotel operators. Maintaining meaningful presence across ChatGPT, Gemini, Perplexity, and Claude requires structured, machine-readable content at a standard that most independent hotel websites do not currently meet. A 2026 Skift analysis found that when AI systems were asked about specific hotel properties, they frequently cited third-party review and comparison sources — including NerdWallet — rather than the hotel’s own website. The hotel with the deeper, more consistent, more machine-readable data footprint will be surfaced more reliably than the hotel with the better product but weaker digital signal.

OTAs, by definition, have the deeper data footprint. Their structured rate-and-availability feeds, review aggregations, and verified content databases are already indexed and retrievable by AI systems. A hotel that relies on OTA distribution is, in effect, also relying on OTA data infrastructure to exist in AI search results.

What the early movers are actually building

A small number of branded hotel groups have begun treating AI search presence as a distribution investment rather than a marketing exercise. Hyatt’s 2025 redesign of its direct booking interface — built around natural language search rather than city-date-availability forms — produced measurably higher conversion rates and has been followed by a branded ChatGPT application launched in February 2026, according to Hospitality.today reporting. The model is consistent: AI surfaces the hotel, the transaction completes on Hyatt’s own platform. That architecture requires significant investment in data infrastructure and brand trust. It is accessible to major chains with loyalty programs and technology resources. Its accessibility to independent operators is less clear.

The Model Context Protocol (MCP), an open connector standard introduced by Anthropic in late 2024 and adopted by OpenAI, Google, and Microsoft in 2025, represents a potential infrastructure path for direct AI-native booking at scale. A startup named DirectBooker — backed by former Tripadvisor CEO Stephen Kaufer and former Google Travel head Richard Holden — is building MCP-based connectivity between hotel booking engines and AI platforms, with active integrations in development with property management systems Eviivo and Mirai. Whether this infrastructure gains sufficient adoption to shift booking economics meaningfully is an open question; the commercial incentive for AI platforms to route transactions through independent booking engines rather than OTAs has not yet been established.

4. The Risk Architecture for Hotel Operators: Four Lines That Warrant Attention

Concentration, data poverty, agentic deferral, regulatory distance

The four structural risks for hotel operators emerging from the 2025–2026 data are distinct and compound each other.

The first is channel concentration. OTA share of independent hotel bookings at 63.4% — with some markets at 80% — means that demand is concentrated in a small number of channels controlled by two companies with the pricing power and data advantage to maintain that concentration. AI-driven discovery is, at present, adding volume to those channels.

The second is data poverty. Every OTA booking is a transaction where the hotel does not own the guest relationship. Contact data, preference history, and behavioural signals flow to the platform rather than to the property. As AI personalisation raises consumer expectations for tailored experiences, hotels without first-party data are at a growing disadvantage in building the direct relationships that protect against commission dependency.

The third is the agentic transition risk. The current AI discovery architecture — where AI recommends and OTA fulfils — could shift if agentic AI systems gain the ability to complete full booking transactions natively, without routing through OTA interfaces. OpenAI’s decision in early March 2026 to retreat from its native ChatGPT checkout functionality, confirmed by the company to Skift and PhocusWire, suggests this transition is slower than initially anticipated. Expedia’s share price rose 12% and Booking Holdings 8% on the announcement, indicating the market had priced in the disintermediation risk and was relieved at its deferral. That relief does not eliminate the risk; it defers it.

The fourth is the regulatory dimension. The EU’s Digital Markets Act creates ongoing obligations for large platform operators regarding data access and pricing transparency, but enforcement timelines and practical effects on OTA commission structures remain unclear. Hotels operating in European markets should monitor DMA developments, but no concrete commission relief has materialised from regulatory action to date.

Renting market position at a faster compound rate

These four risks share a common root: any hotel whose demand generation depends primarily on OTA visibility is renting its market position rather than owning it. That has been true for a decade. The AI transition has not changed the structure; it has changed the speed at which the rental fee compounds. A hotel with 63% OTA share and ADR of €150 is paying approximately €14–17 per room night to the channel, before accounting for the margin erosion from higher cancellation rates and lost upsell revenue.

According to the Cloudbeds report, global occupancy for independent hotels slipped 0.6% year-on-year in 2025, while ADR fell 5.8% and RevPAR declined 5.4%. Branded hotel performance over the same period moved in the opposite direction. The divergence is consistent with the structural disadvantage of properties without loyalty programs and direct booking infrastructure in a distribution environment that is consolidating around two dominant platforms.

The numbers that will tell the story before it is obvious

The signals worth monitoring over the next 12 to 18 months include: the pace at which AI booking penetration grows relative to AI discovery penetration (these are currently very different numbers); whether MCP-based direct booking infrastructure gains meaningful property-level adoption or remains confined to technology-forward chains and select independent groups; and whether either Booking Holdings or Expedia moves to restructure commission terms in response to the AI-era value proposition shift — a scenario that has been discussed in industry commentary but has no current operational evidence behind it.

Hotels that are shifting their distribution mix toward metasearch and direct channels are doing so in a market where the competitive cost of that shift — the risk of lower occupancy — has increased as OTA platforms have deepened their hold on AI discovery. The timing calculus is genuinely difficult. What is not difficult is identifying the direction of travel: the platforms that controlled hotel distribution before AI arrived are, by the evidence of 2025, better positioned to control it after.

Data Sources

Most read

OUR PUBLICATIONS

Also to discover