The Hotel Industry Speaks Fluent AI. It Has No Idea What It Just Said.
On May 8, 2026, Cendyn published an article titled “What ChatGPT’s Decision to Move Away from In-Platform Transactions Means for Hotels.” The headline: “AI is changing hotel search. Direct bookings are back in your hands.”
The argument was that ChatGPT stepping back from in-platform hotel transactions is good news for hotels. Hotels that adapt to GEO and AI-first search will be the ones showing up when high-intent travelers ask where to stay. The article closed with an invitation to contact Cendyn for help getting GEO off the ground and future-proofing hotel digital marketing.
The article was already wrong when it was published. Expedia had been inside ChatGPT for seven months.
Sixteen days later, on Sunday, May 24, 2026, during Memorial Day weekend, OpenAI sent an email to ChatGPT Plus and Pro subscribers announcing that ChatGPT could now use connected Gmail, past chats, saved memories, and personal files to generate personalized travel recommendations. The travel use case was front and center. The example in the announcement: “Based on what you know about me, where should I travel next?”
Expedia had already secured a live inventory presence inside the AI planning interface since October 2025. Dynamic flight and lodging results powered by Expedia, including pricing, availability, and a direct booking path, surface within ChatGPT for a more complete and seamless way to plan. Booking.com was also a named OpenAI travel partner. The booking surface was already inside the AI conversation before most hotels knew the conversation existed.
The industry spoke fluent AI for sixteen days and missed the actual development entirely.
The Cendyn article got it wrong. And it is not alone. This is what the industry’s relationship with AI vocabulary looks like in practice. Skift’s reporting on the OTA and AI discovery shift confirms that travelers are now using AI tools to form opinions, narrow options, and build itineraries before they open a booking site. The consideration set forms before the transaction. Control the consideration set and the transaction follows wherever it goes. Stepping back from the transaction while upstream discovery begins drawing on live OTA inventory is not a retreat. It is a more efficient form of the same structural dynamic the industry has been living inside for twenty years.
The Cendyn article named the correct threat in one sentence – “This changes AI from a booking competitor to a major discovery channel” – and sold a GEO service against the wrong layer in the next.
Richard Feynman Would Have Asked One Question
Richard Feynman’s father taught him early that knowing the name of something and knowing the thing are not the same. A bird on a fence has a name in every language. Know all of them and you still know nothing about the bird. What it eats. How it flies. What happens to it in winter. The name is a social agreement about what to call something. It is not the thing.
Feynman carried that distinction his entire career. The test he applied to everything: can you explain it in plain language, without the vocabulary that substitutes for understanding it? If you can, you know the mechanism. If you reach for the terminology, you know the label.
Apply it to AI.
You know the word GEO. Generative Engine Optimization. Now explain the mechanism. Not “optimizing for AI-generated answers.” That is the label dressed in a longer sentence. Explain what changes inside a language model’s output when the authoritative source density around a brand increases. Explain why a model drawing predominantly from OTA listing copy and review aggregators will describe your property in OTA vocabulary regardless of what your website says.
Modern AI answers come from a mix of training data and live retrieval. The specific weighting depends on the platform, the query, and the retrieval architecture in use. The shared problem across both layers is the same: hotels often do not control enough of the authoritative information environment that AI systems draw from when they describe, rank, and recommend properties. That environment includes OTA listings, review platforms, aggregator summaries, structured data, and brand-owned content, weighted by authority, frequency, and distribution across sources the model can access. A GEO audit measures surface signals. It does not change the composition of that environment. Those are different problems.
Most in the industry cannot explain that distinction. Not because they lack intelligence. Because they learned the label and stopped there.
The label went into the deck. The deck got presented. The budget got allocated. Nobody asked what was underneath it because the word had already done its job of making the room feel like it understood something.
The Vendors Have a Reason Not to Ask It
Here is the part nobody in the trade press is writing.
The market rewards vocabulary-first selling. As long as the industry is debating GEO versus AEO versus AI-first search strategy, the conversation stays at the service layer. It never reaches the structural layer. And at the structural layer is where the questions get uncomfortable.
At the structural layer the question is not which AI visibility vendor to hire. The question is why the model describes your property the way it does, where that description came from, and what would have to change at the information architecture level for the model to draw from something different. That question does not have a product answer. It has a framework answer. And framework answers do not generate recurring revenue.
So the conference panels discuss GEO. The trade press publishes GEO guides. The vendors sell GEO audits. The hotel buys the audit, receives a score, and moves on believing the problem has been addressed. The model still draws from the same information environment it drew from before. The structural condition has not changed.
The Cendyn article is a clean example of how this plays out in real time. The author identified the mechanism accurately, AI has become a discovery channel that shapes consideration before booking begins, and in the same piece sold a GEO service against that mechanism. The discovery layer problem requires an intervention in the information environment itself. The product being sold is a search optimization audit. Those are not the same thing. The article does not explain the difference. That omission is the problem.
The GEO audit was already answering the smaller question before the larger one had been understood.
The industry learned to speak fluent AI the same way it learned to speak fluent direct booking strategy a decade ago. Loud. Confident. With a vendor standing next to them holding an invoice.
What Fluency Without Understanding Actually Costs
The Feynman failure is not an academic problem. It has a specific commercial cost that compounds.
When an industry learns the name of a problem without learning the mechanism, it buys solutions to the label rather than solutions to the condition. The label changes. The condition does not. The industry concludes that the first solution did not work, purchases a new one with updated vocabulary, and repeats the cycle. The condition compounds the entire time.
Hotels did this with OTA dependence. They learned the phrase. They bought direct booking tools, loyalty programs, rate parity software, metasearch management platforms. Each of those tools addressed a real downstream problem: channel cost, conversion, guest retention. None of them addressed the upstream structural condition, which was that the OTA controlled the information relationship between the hotel and the traveler before the traveler had formed a preference. The tools were built for the layer after that point. The structural condition compounded regardless.
AI moves that upstream layer earlier still. The consideration set forms before the traveler starts searching. The model draws from whatever information environment shaped its understanding of your brand, across training and retrieval both. If that environment is dominated by OTA listings, review aggregators, and generic travel content, the model has an OTA-derived understanding of your property. A GEO audit score does not change that. A new AI visibility vendor does not change that. What changes it is a deliberate intervention in the information environment itself, across enough authoritative and independent sources, that the model has a different basis for describing you. That is not a product. That is a structural intervention that requires understanding the mechanism before purchasing anything.
The industry is purchasing. It is not understanding.
The Same Test. The Same Question.
When this test was previously applied to OTA dependence, the conclusion was the same: the industry knew what to call it. The platforms knew what it was worth. The mechanism underneath the phrase was never explained, and that gap cost the industry two decades.
The same calculation is running now. Applied to GEO. Applied to AI-first search. Applied to every conference session, trade press guide, and vendor article that has taught the industry to speak a language it does not understand.
Cendyn published their article on May 8th. OpenAI sent the email on May 24th. Expedia was already in the room.
You know what to call it.
The platform knows what it is worth.
The question is whether the industry is going to learn the mechanism before that calculation completes, or whether it is going to buy another product with the right vocabulary and call it a strategy.
Americas Great Resorts has operated a proprietary affluent traveler demand infrastructure since 1993, built before OTAs existed and 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 at the mechanism level, not the vocabulary level. The full account of what the May 24 OpenAI announcement means for independent luxury hotels is documented at ChatGPT Just Started Recommending Hotels. Expedia Is Already in the Room.
