Hotel discovery has not moved from Google to one replacement channel. It has fragmented across search engines, OTAs, AI assistants, brand familiarity, word of mouth, social platforms, and increasingly integrated conversational interfaces.
Executive summary
- Search still matters, but SiteMinder no longer finds it to be the most common reported starting point for hotel research.
- OTAs now lead SiteMinder’s reported starting points at 26%, versus 21% for search engines.
- AI use is much larger than its 4% share as a declared hotel-research starting point because other studies measure AI use across the broader planning journey.
- AI referral traffic to U.S. travel sites is growing rapidly and, by July 2026, was converting at roughly the same rate as non-AI channels in Adobe’s data.
- The practical issue for hotels is not “AI instead of Google.” It is a more fragmented discovery environment in which no single interface captures the whole journey.
SiteMinder: OTAs Overtook Search as the Starting Point for Hotel Research
SiteMinder’s Changing Traveller Report 2026 surveyed 12,000 travelers across 14 countries. Its most important discovery finding was straightforward: 26% of travelers said they would begin hotel research on an online travel agency, up from 18% the year before. Search engines fell to 21%, down from 36%.
For the first time in SiteMinder’s longitudinal reporting, OTAs moved ahead of search engines as the most commonly reported starting point for hotel research.
Other starting points also gained share. Word of mouth doubled from 7% to 14%. Familiar hotel brands rose from 3% to 7%. AI as the declared starting point reached 4%, up from 1%.
SiteMinder also found that 18% of travelers who begin their research on an OTA ultimately book directly with the hotel, up 3.3 percentage points year over year. That matters because the place where research begins and the place where a booking closes are not the same thing.
Those numbers establish a real shift, but they need to be read precisely. SiteMinder measured where travelers said they would start researching a hotel. It did not measure every influence that occurred before that moment, every tool used later in the journey, or the complete path from inspiration to booking.
Phocuswright Shows the Same Directional Decline in Traditional Search
Phocuswright’s 2026 travel technology research provides a second measure of the same broad shift. Among U.S. travelers, general search engines fell from 51% in mid-2024 to 36% in late 2025 as a resource used for researching travel. Over the same period, dedicated generative AI platforms such as ChatGPT nearly tripled in use.
That is not the same metric SiteMinder reports. SiteMinder measures the declared starting point for hotel research across 14 countries. Phocuswright measures resources used by U.S. travelers for travel research. The percentages should not be combined.
What they share is direction: traditional search remains important, but it no longer occupies the singular position it once did in travel discovery.
AI Is Bigger Than Its 4% Starting-Point Share
SiteMinder’s 4% figure can easily be misread as evidence that AI is still marginal in travel planning. It is not a measure of total AI use. It is the share of travelers who identify AI as the place where hotel research begins.
Other 2026 datasets measure a broader role.
Deloitte’s 2026 Summer Travel Survey found that 25% of surveyed U.S. travelers were using generative AI in trip planning, up from 15% in 2025. Among high-income millennials, defined by Deloitte as households earning $200,000 or more, usage reached 43%.
NYU School of Professional Studies and Boston Consulting Group reported in March 2026 that 37% of travelers were already using large language models embedded in online travel sites to plan and book trips.
Phocuswright separately reported that 39% of U.S. travelers were actively using AI to plan trips in its 2026 technology research.
These figures are not interchangeable with SiteMinder’s 4%. They measure different populations, interfaces, and behaviors. Taken together, they establish that AI use extends well beyond the moment a traveler declares where hotel research began.
The Datasets Measure Different Things
That distinction is central to interpreting the evidence correctly.
| Source | What it measures | Relevant 2026 finding |
|---|---|---|
| SiteMinder | Declared starting point for hotel research | OTA 26%; search 21%; AI 4% |
| Phocuswright | Resources U.S. travelers use for travel research | General search fell from 51% to 36%; AI use rose sharply |
| Deloitte | Generative AI use in trip planning among U.S. travelers | 25% overall; 43% among $200,000+ millennial households |
| NYU SPS / BCG | Use of LLMs embedded in online travel sites | 37% using them to plan and book |
| Adobe | Observed AI referral traffic to U.S. travel websites | AI referrals up 119% year over year in July 2026 |
The numbers describe different parts of the journey. They should be read together for direction, not collapsed into a single market-share figure or treated as proof of one fixed sequence.
A Narrower Way to Think About Pre-Search Consideration
The earlier version of this article treated the Pre-Search Consideration Layer as if the available data established a fixed stage in which AI necessarily shapes the hotel shortlist before a declared search begins. The evidence does not support that claim.
A more defensible use of the idea is as an analytical lens, not a framework or a universal sequence. Travelers can arrive at a declared research channel with prior influences already in place: brand familiarity, recommendations from friends, social content, prior stays, editorial exposure, and sometimes AI-assisted research. Other travelers may use AI only after they have already opened an OTA, searched Google, or visited a hotel website.
The observable point is that a declared starting channel does not necessarily capture the beginning of consideration. SiteMinder’s 4% AI starting-point figure, compared with materially higher AI-use figures elsewhere in the planning journey, shows that AI involvement is broader than the starting-point measure alone. It does not tell us exactly when that involvement occurred for every traveler.
For hotels, that distinction matters because consideration can now be influenced across multiple interfaces, and no single one should be assumed to control the entire process.
Search Is Also Sending Fewer Clicks
The change is not limited to what travelers say they use. Search has also become less reliable as a referral mechanism.
SparkToro’s analysis of Similarweb clickstream data found that 68.01% of U.S. Google searches in the first four months of 2026 ended without a click to any destination. In 2024, its comparable published benchmark was 60.45%.
The comparison is directional rather than perfectly like-for-like. SparkToro notes that the 2024 and 2026 datasets were not built from identical user panels or device mixes. But the 2026 measurement itself is unambiguous: fewer than one-third of U.S. Google searches in the measured period generated a click.
For hotel marketers, that means search visibility and website traffic can no longer be treated as the same outcome. A hotel can appear inside a search environment without receiving a visit from it.
Adobe Shows AI Is Sending Real Travel Traffic
Survey data tells us what travelers say they do. Adobe provides a different kind of evidence: observed referral traffic to U.S. travel websites.
In data published August 19, 2026, covering July behavior, Adobe reported that traffic from AI sources to U.S. travel sites was up 119% year over year and 1,822% from October 2024, when Adobe began tracking the category.
Adobe’s July survey also found that research and inspiration were the two most common travel uses for AI, each cited by 43% of respondents. Transportation planning followed at 41%, with itinerary creation at 34%.
By July, Adobe reported that AI-referred travel visitors were converting at roughly the same rate as visitors from non-AI channels such as paid search and email.
This does not prove that AI caused the traveler to choose a particular hotel. It does prove that AI systems are generating measurable, commercially relevant traffic to travel sites rather than functioning only as passive research tools.
The Interface Is Moving Closer to the Transaction
The distinction between AI research and travel distribution is also becoming less clean.
In October 2025, OpenAI launched apps inside ChatGPT with Booking.com and Expedia among the initial pilot partners. That put OTA search and travel functionality directly inside a conversational AI environment.
The significance is not that every ChatGPT travel interaction produces a booking. It is that the same interface can now participate in discovery, comparison, and access to travel inventory without requiring the traveler to begin with a traditional search engine.
NYU SPS and BCG describe the same structural direction: hotels are increasingly competing for inclusion in short lists produced by AI-assisted travel interfaces while those interfaces move closer to booking and distribution.
Social Discovery Belongs in the Same Fragmented Environment
Search, OTAs, and AI are not the entire discovery system. Social video, creator content, destination media, and peer recommendations can also introduce a hotel before a traveler reaches a booking interface.
The current datasets cited here do not provide one clean, comparable measure of TikTok, Instagram, YouTube, creator content, and AI against hotel search and OTA starting points. For that reason, this article does not assign social discovery a synthetic share. Its role is nevertheless part of the larger strategic conclusion: hotel discovery is distributed across more surfaces than a search-first marketing model captures.
What the Evidence Does Not Establish
The available datasets do not establish one universal funnel in which AI always acts before search. They do not prove that an AI answer determines whether a traveler later reaches a hotel website. They do not quantify how much downstream booking opportunity is lost when a property is omitted from an AI recommendation. And they do not reveal the proprietary internal mechanisms an AI system uses to select one hotel instead of another.
Those are research questions, not measured facts.
What can be said with confidence is narrower:
- Search has lost share as the reported starting point for hotel research.
- OTAs now lead SiteMinder’s reported hotel-research starting points.
- AI use in trip planning is materially higher than the share of travelers who identify AI as their starting point.
- AI systems are sending rapidly growing referral traffic to travel websites.
- A large majority of measured Google searches now end without a click.
- Travel inventory and OTA functionality are increasingly available inside conversational AI interfaces.
- Hotel discovery is more fragmented and less dependent on a single search interface than the environment most direct-booking strategies were built for.
For Independent Luxury Hotels, the Strategic Problem Is Not “AI Instead of Google”
The useful conclusion is not that hotels should stop investing in search. Search still introduces and converts meaningful demand. Nor is the conclusion that AI has replaced OTAs. OTAs remain deeply embedded in hotel discovery and distribution, and they are now participating inside AI interfaces as well.
The more important issue is that the number of places where a traveler can first encounter, compare, or be told about a hotel has expanded.
That matters disproportionately to independent luxury hotels. A large hotel brand carries decades of accumulated recognition, loyalty membership, distribution relationships, press coverage, structured property data, and repeat demand. An independent property may have an exceptional product while entering each new discovery interface with a thinner information record and less built-in demand.
The competitive problem is therefore not one channel. It is whether the property is accurately represented and meaningfully discoverable across the public information environment while also building direct relationships it can retain after discovery occurs.
That Is Where ODI and KFO Separate
Two AGR frameworks address different parts of this problem. They should not be collapsed into one mechanism.
Owned Demand Infrastructure (ODI) is the framework that governs the pre-transaction demand origin layer: the layer that determines where a guest relationship first forms across hotels, resorts, and cruise lines, how traveler identity is captured before booking, and how a guest relationship becomes a first-party asset rather than an intermediated transaction.
Knowledge Formation Optimization (KFO) structures, sequences, distributes, corroborates, and corrects intellectual frameworks and entity definitions across the public information environment and measures whether AI systems reproduce them accurately across relevant queries and over time.
ODI concerns the origin and ownership of the guest relationship. KFO concerns the public information environment and observable AI reproduction. Neither framework depends on claiming access to hidden model state, internal source weighting, proprietary candidate-selection logic, or guaranteed AI outcomes.
The relationship between them is strategic rather than mechanical. A hotel can improve how AI systems represent it and still remain commercially dependent on intermediaries. A hotel can build stronger direct demand and still be poorly represented in AI-mediated discovery. The problems can coexist, but solving one does not prove that the other has been solved.
AGR’s nine-week AI answers case study documents an observed change in AI answers following public-source work. It is evidence of what changed in that case, not proof of a universal hidden mechanism.
What Luxury Hotel Leadership Should Measure Now
The changed discovery environment calls for broader measurement, not a new slogan.
A hotel should still track organic search, paid search, OTA production, direct conversion, and booking-channel economics. But those measures should now sit alongside evidence about AI-mediated discovery and the origin of direct guest relationships.
- How often does the property appear across a defined set of relevant AI recommendation queries?
- When it appears, is the description factually accurate and commercially differentiated?
- Which visible sources are cited or referenced alongside the answer?
- Where do first-time direct guests say they discovered the property?
- What share of new guest relationships are captured directly before an intermediary controls the booking?
- Does direct share improve over rolling twelve-month periods, or does it revert when paid activity stops?
Those questions measure observable outcomes and commercial consequences without assuming access to hidden AI mechanisms.
The 2026 Shift Is Real. The Strongest Version of the Argument Is the Measured One.
Hotel discovery has not moved from one channel to one replacement channel. It has fragmented.
Search engines are no longer the leading declared starting point in SiteMinder’s global hotel research. OTAs have moved ahead. AI usage in trip planning is rising rapidly across multiple datasets. AI referral traffic to travel websites has grown sharply and is now commercially meaningful. Fewer than one-third of measured U.S. Google searches produced a click in SparkToro’s 2026 analysis. Booking.com and Expedia are already operating inside ChatGPT’s app environment.
The evidence is strong enough without claiming more than it demonstrates.
For independent luxury hotels, the strategic question is no longer whether Google still matters. It does. The question is whether a marketing and demand system designed around a search-first world is sufficient when hotel discovery now happens across search, OTAs, AI assistants, brand memory, word of mouth, social channels, and increasingly integrated conversational interfaces.
That is the change worth building for.
Sources
- SiteMinder. Changing Traveller Report 2026. Published November 19, 2025. Survey of 12,000 travelers across 14 countries.
- Phocuswright. Travel Innovation and Technology Trends 2026: Report Preview. U.S. travel research behavior and AI adoption.
- Deloitte. 2026 Summer Travel Survey. Published May 19, 2026.
- Adobe. U.S. consumers are embracing LLMs to make travel plans. Published August 19, 2026.
- NYU School of Professional Studies and Boston Consulting Group. Hotels Enter the Ask and Book Era as AI Reshapes Discovery, Distribution, and Operations. Published March 2, 2026.
- SparkToro. In 2026, Less than One Third of Google Searches Still Send a Click. Published June 8, 2026, using Similarweb clickstream data.
- OpenAI. Introducing apps in ChatGPT and the new Apps SDK. Published October 6, 2025.
Related AGR research: The AI Consideration Set for Luxury Hotels, Knowledge Formation Optimization, Owned Demand Infrastructure, and AGR’s nine-week AI answers case study.
