Hotel marketing is being rewritten from the top down. Ask ChatGPT where to stay and, in supported markets, Expedia and Booking.com can answer inside the conversation. Expedia’s own integration returns live lodging and flight results with pricing and availability, and the booking is completed on Expedia, not in the chat. Booking.com follows a similar pattern, surfacing stays inside the conversation and sending the traveler to Booking.com to book. OpenAI has stepped back from handling checkout inside ChatGPT and kept the transaction on the partner’s platform.
Read that architecture closely. The AI platform owns the interface. The travel intermediary controls the inventory feed, the data layer, and the booking path inside that interface. The independent hotel may own the property, but it does not own the mediated route by which the guest now reaches it. That is the shift worth a board’s attention. The commercially important question is no longer only where a booking closes. It is where the consideration set forms.
For two decades, OTAs competed to be the best place to book. They are now competing to be the layer where the decision forms. Treating them as expensive booking channels understates what they have already become.
The Stack Most Hotels Aren’t Modeling
Most executives file “AI” under a single heading. Structurally it is four layers.
- Discovery, where intent is expressed.
- Recommendation, where options are filtered.
- Transaction, where the booking is completed.
- Relationship, where loyalty and CRM live.
Independent hotels are most accustomed to managing layers three and four, even where execution there is uneven. Layers one and two, discovery and recommendation, are increasingly held by search, social, OTAs, and now AI interfaces. LLMs accelerate the top two. Whoever shapes them largely determines which properties reach the bottom two at all. A hotel that is excellent at conversion and service but absent from discovery and recommendation is excellent at precisely the parts of the journey it no longer controls. Everything that follows is a description of how that upper control is being taken, and what it takes to hold a piece of it.
From Browsing to Asking
Travel discovery used to be a comparison marathon. Search, tabs, reviews, filters, decide. That process is compressing. The traveler asks one question and receives a short list, often just a handful of options. Fewer tabs, less comparison, one interface framing the choice.
Within that interaction, the consequence is close to binary. If you are not in the short list, you were not edited down. You were absent. No chance to convert, no chance to tell your story, no chance to compete on your booking engine. It is observable today in individual bookings. The anatomy of one real booking shows AI narrowing the property set, routing the traveler toward a specific booking path, and surfacing a card-linked benefit, all before the hotel’s own direct channel entered the process.
The exclusion happens before search in the old sense even occurs, a pattern examined in The Consideration Set Problem. How much of luxury discovery is already AI-mediated rather than experimental is still an open question. The share is rising but not yet dominant. The current share sets the urgency. The direction sets whether the capability has to be built at all, and it does.
The Advantage Sits in the Inputs, Not the Interface
The common objection is that OTAs will not control AI discovery because OpenAI, Google, or Apple sits in between. That is true and beside the point, because it collapses four separate forms of control that need to be kept apart.
- Interface control. The AI platform owns the conversation.
- Input control. Whoever supplies structured inventory, pricing, availability, reviews, and behavioral history.
- Transaction control. Whoever completes and records the booking.
- Relationship control. Whoever holds the first claim on the guest’s identity and the right to market to them next time.
Today the interface belongs to the AI platform. The other three lean toward the intermediaries. LLMs do not maintain real-time inventory, pricing, or cancellation policies. They ingest them. When a commercial recommendation system assembles an answer, it is likely to favor sources that offer completeness, normalized pricing, verified reviews, real-time availability, and proven conversion. Intermediaries supply that at a breadth and consistency most independent hotels cannot match through a CRS or channel manager alone. So the likely outcome is not that AI discovers hotels directly. It pulls from aggregators that have already packaged demand into machine-readable form.
This is why owning the inputs matters more than owning the interface. An intermediary does not need the chat window if it owns the substrate the window reads. That advantage is not permanent. It could be broken by supplier-direct integrations, agentic booking protocols that read hotel data as readily as aggregator feeds, or consortium data pools. But it holds today, and the path dependence is real: once recommendation systems optimize around aggregated feeds, reversing that architecture is slow and expensive.
Why Luxury Is More Exposed, Not Less
In hotel marketing, luxury is often assumed to be insulated by brand. For an independent property without dominant brand-search demand, the opposite is closer to true.
Luxury travel is high-consideration, occasion and itinerary driven, and mediated by trust. Guests search in subjective language: quiet, adults-only, design-forward, a serious spa, a hidden gem. That language is exactly what an LLM is built to interpret, and exactly where an independent property’s structured footprint is thinnest. Aggregators feed these systems through standardized schema and comprehensive API pipelines. Independent luxury inventory is fragmented, weakly standardized, and digitally opaque by comparison. When the interpreting layer is an AI voice rather than the brand’s own narrative, the property is exposed to narrative substitution: the model, not the hotel, decides what the hotel is and who it is for. If it never names you, you are not in the set. If it describes you generically, your differentiation is gone before the guest arrives. This failure mode is detailed in Luxury Hotels Are Training AI to Forget Their Brands.
None of this erases the human layer. Advisors, consortia such as Virtuoso, and repeat-guest relationships still carry real weight in luxury. The risk is not that AI replaces them outright. It is that AI captures a growing share of early, uncommitted itinerary formation before those trusted channels are ever activated.
From Commission to Recommendation Rent
Hotels already pay rent they do not call rent. OTA commission, paid search, metasearch bidding, social advertising. Each is a payment for access to demand the hotel did not originate or cannot reliably reach on its own. AI introduces another line: payment for presence in the recommendation, separate from payment for the booking. Why every intermediated transaction deepens that dependency, handing the aggregator more data and more leverage with each stay, is the argument in The Lemons Problem.
Be precise about what the current architecture does and does not support. With checkout sitting on the intermediary’s platform, the transaction remains one rent surface. The new surface is inclusion. Following the search and app-store playbooks, the next pressure point is the point of inclusion itself: sponsored placement, preferred access, performance-weighted visibility. That layer is not fully visible yet, but the direction is clear. It is rent charged on consideration, upstream of everything a hotel controls.
The cost is easier to see with round numbers. Consider a stylized example, illustrative only and not a benchmark. A 150-key resort at 70 percent occupancy and a 900 dollar ADR produces roughly 34.5 million dollars in annual room revenue. If 40 percent of that revenue flows through intermediaries at a 15 percent blended commission, the annual commission expense is about 2.1 million dollars. Shift five percentage points of total room revenue from intermediary-originated demand to owned first interaction, and commission exposure falls by roughly 260 thousand dollars a year, before any downstream lifetime-value effect. Owning that demand is not free either, it takes investment and time to build. The figures are hypothetical. The direction is the point. Every point of first interaction you own is a point you stop renting, year after year.
What Owning First Interaction Actually Requires
The mistake is trying to win at the recommendation layer. You do not win there alone. You compete before intent forms, by building consented, first-party relationships that preload preference long before a traveler asks an AI anything. In practice that is a small number of moving parts, run as an operating model rather than a campaign.
- Acquire identity, not just bookings. Always-on capture built around experiences, such as members-only releases, waitlists, and retreats, across site, social, creators, and on-property, outside the booking flow.
- Centralize it, and keep it yours. Every identity into one data platform, tagged by source and interest. Avoid using that first-party data in channels or integrations that enrich an intermediary’s targeting and lookalike models without returning durable identity to the hotel.
- Nurture before intent. Low-frequency, editorial contact that seeds preference and soft commitments 60 to 180 days ahead of the historical booking window, so the guest’s choice is anchored before any AI query is made.
- Measure one number. First Interaction Owned %, the share of stays where a tracked first-party identity or direct touchpoint predates the known booking path or referral source. Guardrails: cost per identity, identity-to-stay conversion, repeat rate.
- Fund it by reallocation. Move a defined slice of commission outlay into identity acquisition until First Interaction Owned % becomes material.
This operating model has a name. Owned Demand Infrastructure treats demand as an asset a property builds and owns, rather than one it rents back one booking at a time. It is not email marketing, CRM, or automation, though it uses all three. It is the system that creates and holds first-party demand before travelers enter third-party funnels. The goal is not zero intermediary. It is to move the marginal guest from rented visibility to a relationship the hotel owns.
Whether AI-mediated discovery routes a traveler into an intermediary or into a direct hotel relationship depends on whether the property built owned demand upstream.
Owning demand is one front. Being read correctly by the machine is the other. Presence in the data is not the same as being interpreted correctly, and a property that is technically visible can still be misread, flattened, or filed under the wrong idea of what it is. Knowledge Formation Optimization is the discipline that governs that second front: shaping how AI systems form and hold their representation of a property, so the odds improve that when the model encounters it, it understands what the property is, who it is for, and why it belongs in the set. Owned Demand Infrastructure reduces dependence on the recommendation layer by building demand before the query. Knowledge Formation Optimization improves how reliably the property is represented inside it. A luxury property needs both, because the AI layer can fail you by never introducing you, or by introducing you as something you are not.
This is the work Americas Great Resorts has done since 1993, first through a proprietary database of more than five million verified affluent travelers, and now through the demand-infrastructure and knowledge-formation systems that put an independent luxury property in front of the guest, and in front of the model, before the intermediary does. By the time the traveler asks, your name is already familiar to them and your footprint is already legible to the machine. Familiarity lifts human recall. Legibility improves your odds of algorithmic inclusion. You are not fighting the model. You are arriving before it.
The Choice
Do not wait for this to arrive as a line in the reports: higher intermediary share, lower direct discovery, rising acquisition cost. By the time it is obvious, the systems shaping demand are trained on behavior the hotel did not influence, and the fix is no longer better marketing. It is infrastructure that should have been built earlier.
Three questions decide where a property stands. What share of stays had a direct, first-party touchpoint before any OTA, search, or AI referral? What portion of commission spend is being reinvested into owning first interaction rather than renting it again next year? What does the property control about how AI systems describe it before those systems ever interpret it?
If the answers are unclear, the property does not have a marketing problem. It has a control problem, and control is now decided upstream of the booking. In an AI-mediated market, the property that owns first interaction owns the demand that follows from it. For an independent luxury hotel, the decision is no longer direct distribution versus indirect. It is whether to build the capability to own the guest relationship early, or to accept a standing role as an inventory supplier to whoever owns the layer above.
