“I’m sorry Dave, I’m afraid I can’t do that.”
In 1968, Stanley Kubrick introduced the world to HAL 9000, a perfectly polite, enormously capable artificial intelligence working entirely in its own interest while the crew depended on it for survival. HAL did not malfunction. He optimized. The crew simply did not understand what he was optimizing for.
That scene is no longer fiction. It is your technology vendor relationship.
And you invited him in.
Every Platform That Ever Helped You Was Watching You
The history of technology platforms follows a recurring pattern so consistent that executives who have lived through it once should refuse to be surprised by it twice.
Act One: openness. The platform needs builders, developers, and dependents to establish market position. It distributes tools, subsidizes access, and makes itself indispensable. The cost is low. The terms are generous. The promise is partnership.
Act Two: observation. The platform watches what gets built on top of it. It tracks usage patterns, identifies winners, and maps every dollar of value being created by others on its infrastructure. The data does not go to waste.
Act Three: extraction. The platform builds competing first-party products, adjusts its algorithm to favor its own offerings, raises prices once switching costs are high enough, or acquires what it cannot replicate fast enough. The partners who built the platform’s value discover they were the product, not the customer. They were never the customer.
The mechanism is not identical in every case. Amazon used marketplace sales data to identify profitable categories, then launched private-label competitors directly into those categories, undercutting the sellers whose data revealed the opportunity. Facebook trained publishers to rebuild editorial operations around Facebook referral traffic, then collapsed organic reach once advertising revenue made the swap worthwhile, a dynamic documented extensively by the Reuters Institute for the Study of Journalism in its annual Digital News Reports. Apple built an app economy on the labor of independent developers, then used App Store governance and a 30% distribution toll to control it entirely on Apple’s terms, a toll structure now under antitrust scrutiny in the U.S., EU, and UK simultaneously. Google converted search visibility into a pay-to-play auction layer, then began replacing organic results with AI-generated answers that keep the user inside Google’s ecosystem rather than routing them anywhere else. Different mechanics. The same recurring structural outcome: once the platform controls access to demand, the dependent party negotiates from weakness.
Hotels already lived through one full cycle of this. It was called the OTA era. Expedia and Booking.com presented as distribution partners. They needed inventory. So they made it easy to list, easy to reach travelers, easy to close bookings. Then commission rates climbed from the mid-single digits in the late 1990s to a documented industry average of 15 to 25 percent by the mid-2010s, according to research published by Cornell’s Center for Hospitality Research. Then ranking algorithms shifted toward properties with higher review volume and conversion rates, systematically rewarding OTA-dependent behavior. Then rate parity clauses appeared in distribution contracts, limiting hotels’ ability to offer lower prices on their own websites, with those clauses subsequently restricted or banned outright in France, Germany, Austria, Italy, and Sweden after years of antitrust investigation by the European Commission. Then the guest relationship belonged. The name, the email, the stay history, the preference data, the lifetime value to the platform. The hotel became inventory. The OTA became the brand. The mechanism of how OTA dependence actually works follows a logic most hotel executives have never fully mapped, and that unexamined logic is exactly what is repeating now.
That cycle took the better part of two decades to complete. Hotels are still paying the bill. Every single day.
The New Platform Is Not a Booking Channel
Here is where luxury hotel executives make the catastrophic mistake of thinking the lesson has been learned.
The OTA lesson, as most executives internalized it: be careful with distribution channels that sit between you and your guest. That is a real lesson. It is also completely the wrong frame for what is happening now.
AI platforms are not distribution channels. They do not sit between you and your guest at the point of booking. They sit between your guest and the question they ask before they ever start searching.
When a qualified traveler opens an AI assistant and asks where to stay in a specific market, the AI does not return ten blue links and let the traveler decide. It forms a considered response based on what it has learned and what its training corpus has weighted as authoritative. The consideration set is built before the traveler touches any booking flow. AI excludes hotels before search begins, not at the point of booking, not at the point of comparison, but at the point of question formation. Hotels not present in that consideration set do not lose a booking. They never enter the decision.
Some luxury operators will argue that their guests do not use AI to choose resorts, arguing that the ultra-affluent traveler relies on elite human networks, private concierges, and specialized advisors rather than digital tools. This is true, and it is also irrelevant. Those advisors are increasingly using AI infrastructure behind the scenes to research properties, synthesize options, and validate recommendations for their clients. If your brand is flattened into a category average inside the model, you do not just disappear from consumer-facing queries. You disappear from the workflow of the human gatekeepers your business depends on for survival.
That is a different category of platform risk than anything this industry has previously managed. Prior platforms controlled where buyers went. This platform controls what buyers think before they go anywhere. Luxury hospitality is entering the post-search era, and most properties are still optimizing for the era that just ended.
HAL did not refuse Dave’s request through a booking interface. He simply closed the pod bay door. There was no channel to optimize. There was no rate to adjust. The decision had already been made, upstream, at the system level. Dave never had a chance to compete. The competition was over before it started.
The Largest Involuntary Intelligence Operation in Human History
Here is what nobody in the industry is saying out loud.
AI platforms are running the largest involuntary intelligence gathering operation in human history. Billions of conversations every day. Every industry. Every geography. Every discipline. The most capable people on earth, self-selected by the fact that they are the ones reaching for AI tools to solve hard problems, walking the platform through their best thinking in real time. Business models. Pricing strategies. Product architectures. Scientific hypotheses. Creative frameworks. Competitive responses. All of it flowing into the same system. All of it processed, retained, and owned by the platform under terms the platform wrote.
This is not passive data collection. Amazon watched what sold. Facebook watched what got shared. Google watched what got clicked. That was behavioral signal, aggregated and statistical. AI platforms watch what people think. Before those thoughts become decisions. Before they become announcements. Before they become products. Before they become patents.
Consider what that means. A medical research scientist has spent three years developing a hypothesis around a novel mechanism for treating a specific cancer. The research is real. The thinking is original. It does not exist anywhere else. Before publishing, before filing, before telling anyone, she uses an AI platform to stress-test the logic, pressure-test the methodology, validate the framework. The platform now has it. The complete hypothesis. The mechanism. The methodology. The language she used to describe what she found. And the platform has something she does not: unlimited capital, unlimited development infrastructure, and zero legal obligation to tell her what happens next. Six months later something that looks like her idea exists in the world. There is no record of the transaction. There is no notification. There is no invoice. She used a tool. The platform walked away with three years of her best work, and there is no court in the world currently equipped to hear the case.
That is not a data privacy issue. That is the most efficient theft mechanism ever constructed. And it is operating across every industry simultaneously, at a scale no intelligence operation in human history has ever achieved, funded entirely by the people being harvested.
For hotel executives the individual stakes are lower. The mechanism is identical. Every strategic question entered into a consumer AI tool, a new pricing model, a demand restructuring concept, a positioning framework that does not yet exist in the market, a competitive response that has not yet been executed, is a voluntary intelligence contribution to a platform that is under no structural obligation to forget it. You are not the customer. You are the research department. You are just not on the org chart.
The bully did not ask for your lunch money. He just watched where you kept it.
The Three-Act Play Is Not Starting From Zero
The AI landscape is not uniform, and that matters strategically.
New entrants, OpenAI, Anthropic, Perplexity, are in genuine Act One. The tools are cheap. Access is generous. The narrative is partnership and enablement. They are building dependency before they build revenue models. Enjoy it. It will not last.
Legacy platforms are doing something more dangerous. Google is not in Act One of anything. In August 2024, a federal court ruled that Google had illegally monopolized the U.S. general search market, the infrastructure it is now using to port its extraction architecture directly into AI interfaces. AI Overviews surface synthesized answers that satisfy user queries without routing travelers to hotel websites. The platform risk is not coming for Google. It arrived. It is being repackaged in a new wrapper and presented to the industry as innovation. The data on hotel travelers leaving Google for AI discovery is no longer a projection. It is the current operating environment.
Both dynamics converge on the same outcome for hotels: the consideration set forms inside a platform layer the hotel did not build, does not own, and cannot negotiate with.
This is the moment Zynga executives thought was permanent. This is the moment independent Amazon Marketplace sellers thought they had found their distribution solution. The recurring platform pattern is control access first, extract value later. Early openness is not evidence of permanent neutrality. It is the cost of acquisition.
The extraction mechanics for AI will not look like commission rates. Perplexity tested a sponsored answers model in 2024, inserting paid placements directly into AI-generated responses, establishing the template for answer-layer monetization that the rest of the industry is watching. Google has already integrated paid hotel listings into AI-driven search interfaces. The trajectory: preferential visibility for brands that purchase platform access; training advantages for properties with structured machine-readable identity; and the gradual disappearance of brands the corpus never learned to describe as anything distinct, does not require a policy announcement. It requires patience and a proprietary dataset. Both of which they have in abundance.
What Gets Controlled Gets Commoditized
The structural principle underneath every platform cycle is the same. The layer controlled by an external platform is the layer that gets commoditized for every company dependent on it.
OTAs controlled distribution. Distribution became a cost center. Hotels that could not own their demand channel competed on price inside someone else’s interface. The product quality that should have been a differentiator was flattened into a star rating and a review score. Decades of brand investment, compressed into five stars and a “Wonderful” badge.
If AI controls preference formation, the same compression happens one layer upstream. The brand narrative that should differentiate a property gets processed through a training corpus where every luxury hotel is described in roughly the same vocabulary. The distinctiveness built over decades of editorial coverage, word of mouth, and direct guest relationship gets averaged into a composite output the AI considers “luxury accommodation in that market.” Indistinguishable from every other property in the comp set. Which is precisely the point. Luxury hotels are training AI to forget their brands, not through any deliberate action, but through the simple failure to shape what the machine learns before the corpus is set.
This is not a metaphor. It is how language models work. A model generating a response about a luxury hotel in a given destination defaults to the highest-frequency semantic associations in its training data. When that training data is dominated by OTA listing copy, generic travel content, and review aggregators, which for most properties it is, the output reflects that corpus. The property that spent thirty years building a singular identity gets described in the same vocabulary as the property that opened eighteen months ago. Your brand does not disappear because the platform dislikes you. It disappears because the platform learned who you are from Booking.com and TripAdvisor, and that is exactly how you sound.
HAL did not hate Dave. HAL had no concept of Dave as an individual. Dave was a crew member with a functional role. When that role conflicted with the mission, the calculation was straightforward. There was no malice. There was no negotiation. There was only the mission.
The mission was never about Dave.
The Structural Answer
The brands that navigated the OTA cycle without full capitulation shared one characteristic. They built demand relationships that existed outside the OTA infrastructure before dependency was complete. Four Seasons, Aman, and the strongest independent properties maintained direct guest relationships, proprietary databases, and acquisition channels that did not route through the platform. When the terms changed, they had somewhere to stand. The hotels that had handed the platform their entire commercial relationship discovered that dependency has a price, and the platform sets it. Luxury hotel demand ownership is not a marketing strategy. It is the only structural defense against a platform that controls access to your future guests.
The structural answer to AI platform risk follows the same logic one layer earlier. It is not about refusing to participate in AI systems. It is about ensuring the machine learns your brand on your terms before the market consolidates around a training corpus you did not shape and cannot correct after the fact.
That is not a technology problem. It is an information architecture problem. The question is not whether your property has reviews on platforms that AI systems index. The question is whether AI systems have encountered enough authoritative, independent, non-OTA-derived information about your brand, distributed across enough distinct and credible sources, that they have a genuine basis for describing you accurately and distinctively when a qualified traveler, or their advisor, asks. Knowledge Formation Optimization is the publishing discipline that governs how that authoritative information gets built, distributed, and indexed before the consideration set is formed.
Most properties do not. Most will not act until the evidence of displacement is visible in their booking data, which means acting after the corpus has already been formed, after the consideration sets have already been shaped, after the competitive disadvantage is structural rather than correctable. The diagnostic question every luxury hotel needs to answer is not whether AI is a risk. It is whether the machine already knows who you are, or whether it learned you from someone else’s description.
The pod bay door is still open. For now.
The question is whether you are going to do anything about it before HAL decides the mission parameters have changed.
Americas Great Resorts has operated a proprietary affluent traveler demand infrastructure since 1993, built before OTAs existed, structured so that no platform intermediates the relationship between AGR clients and their future guests. The Knowledge Formation Optimization (KFO) framework governs how AGR clients shape AI understanding of their brands before the consideration set is formed.
