A hotel can be excellent, bookable, and completely invisible at the same time

In the AI era the binding constraint was never capability. It was sight — over where demand begins, and over what the technology actually returns

For a year now the industry has asked one question about artificial intelligence, in a hundred forms. What can it do. Can it plan the trip, draft the reply, price the room, run the front desk, answer the call at two in the morning. Every conference stage, every launch, every panel has been a variation on capability — how much the machine can take on, how soon, how well.

Capability was never the binding constraint.

The constraint, arriving quietly underneath all of it, is sight. A hotel is losing its ability to see two things it long took for granted: where its demand comes from, and what its own technology returns. Both are moving onto layers the hotel neither owns nor fully observes. The capability race is real. It is also running on top of a problem it keeps the industry from noticing.

The hotel that cannot be seen

Begin with demand. For most of the digital era a hotel could reason about how a guest found it — search, metasearch, an OTA listing, a brand site, an ad. Channels you could name, count, and roughly trace. That map is dissolving.

At NYU's hospitality investment forum this month, Hilton's chief information officer told the room the company has to stay close to Google, ChatGPT, OpenAI, and Anthropic, because no single model will win and the distribution landscape will be unrecognizable within a few years. That is one of the largest hotel companies in the world conceding that the first moment of discovery — the question a traveler asks before any hotel name enters their head — now happens somewhere it does not control. Accor has already placed a booking app inside ChatGPT. The major brands are negotiating terms of presence inside a moment they no longer expect to own.

The same week, American Express agreed to pay $700 million for TheFork, its third restaurant-reservation platform after Resy and Tock. A card network is buying the layer where diners decide, because the company that holds the payment increasingly wants to hold the choice in front of it. Dining is the proving ground. Amex has announced nothing for lodging — but owning the choice, not only the payment, was never a restaurant-specific idea.

And the entry point is narrowing. A Cornell Center for Hospitality Research study, conducted with the marketing-technology firm Curacity, found that more than nine in ten hotels are effectively invisible in AI search — present or absent, with little in between. Whether a property appears is settled by things that have nothing to do with hospitality: how its data is shaped, how fast its pages load, whether an agent can read and price what it sells. A hotel can be excellent and unreadable at once. To the system building the traveler's shortlist, unreadable is the same as absent.

Each of these is the same development in different dress. The place where demand begins is moving onto surfaces the hotel cannot see into.

The hotel that cannot see

Now turn the lens around. While the hotel struggles to be seen, it is also losing the ability to see itself — to know what its own technology actually returns.

IHG offers the cleanest case. Its chief executive described an eight-year rebuild of the company's core systems and said the payoff is reaching franchisee margins through sharper pricing and personalization at scale. Owners pressing for fee relief are being told the technology is the relief. Asked for the figure — the RevPAR premium the new systems produce — he pointed to it and would not put a target on it. The gain is described across 7,000 hotels. The fee is paid one hotel at a time. The portfolio math is asserted; the property-level proof is not on the table.

That gap is not an IHG quirk. It was the loudest theme at this month's Skift Data + AI summit, where two dozen travel technology leaders kept circling the same admission. Cost savings are easy to measure — fewer calls, faster work. Revenue is the hard case. One operator put it flatly: cost can always be measured; revenue almost never can. "AI bookings" get counted with no shared definition of what an AI booking is. Pilots generate learning and postpone the verdict. The instruments that would tell a hotel whether the spend paid — clean attribution, an isolated variable, a property-level line — are the things the industry has not built.

So the hotel pays for capability it cannot yet price, against value it cannot yet see.

One loss, not two

These look like two separate problems. A demand problem for the marketing team, a measurement problem for finance. They are one problem wearing two coats.

In both, the hotel is losing its instruments. Demand is migrating to layers it cannot observe — models, agents, a payments network. Value is migrating to a scale it cannot read from where it stands — the portfolio, the platform, the vendor's dashboard. Outward, the hotel can no longer see where the guest came from. Inward, it can no longer see what its own systems returned. The same erosion, pointed in two directions — and the capability conversation runs straight past it, because capability is the one thing easy to demonstrate on a stage.

A booking still arrives. A dashboard still fills. The hotel knows less and less about how either happened.

What sight would require

The operators worth watching over the next few years are the ones rebuilding their instruments of sight. The work is unglamorous and within reach.

Knowing whether your property appears in an agent's answer, and closing the gap when it doesn't. Structuring your content and data so a machine can read, price, and book what you sell. Insisting, before the next investment, on a way to isolate what a system actually changed at the property — staffing, conversion, satisfaction, cost — instead of accepting a portfolio average as proof. None of it requires owning a model or outspending a chain, only the discipline to treat sight as the asset and to refuse capability that arrives without it.

The industry has answered what AI can do for a hotel, many times over. The question that decides the next few years is quieter, and a hotel can still answer it for itself: can you see your own demand, and can you measure your own return. The properties that can will compete from solid ground. The rest will run a faster and faster race with their eyes closed.

by Markus Busch, Editor and Publisher of Hospitality.today

Enjoying this analysis? Hospitality.today delivers daily insights on hotel distribution, AI trends, and travel commerce — straight to your inbox. Subscribe for free at Hospitality.today →