Poor Hotel Data Is Killing Direct Bookings. C.U.P.S. Can Fix It

For months, I’ve been running the same experiment with hotel managers around the world. I ask: “Have you ever tried asking ChatGPT or Gemini to recommend a hotel in your city?” Almost no one has. So I say: “Try it. Now.”

The result is almost always the same. The AI returns three or four hotels. Theirs isn’t on the list. Or if it is, the price is wrong, it mentions a spa that closed two years ago, or it lists outdated restaurant hours. Almost all hotels have a beautiful website, hundreds of blog posts, and great reviews, but to anyone searching through ChatGPT, Gemini, or Perplexity, it might as well not exist.

This is the early signal of a shift that, within two or three years, will decide which hotels keep capturing direct bookings and which ones will pay ever-higher prices to win them back.

Search no longer leads to links. It delivers answers.

For twenty-five years, hotel distribution was a fight for position on a results page: be at the top, beat the competition, outrank the OTAs. That fight is fading fast. Most searches today end without a single click. The AI-generated answer sits above everything, offering a recommendation before the traveler even considers visiting a website.

The consequence: the top of your sales funnel now closes with one answer already filtered, already decided. In the best case, the customer arrives on your site after the decision has been made elsewhere, without you.

Two internets now coexist. The first is the web of humans: pages, photos, and traditional SEO. The second is the web of agents, where ChatGPT, Google, and personal assistants act as distributors themselves, and recommend only what they can reliably read.

The center of gravity is shifting fast toward the second.

OTAs and chains already won the data race

What do LLMs use to generate answers? Structured data. Not marketing copy, but clean, standardized, machine-readable information: room type, real-time availability, amenities, conditions, exact location.

The major OTAs and large hotel chains have been structuring this data for two decades. The two distributors that dominate global hotel bookings didn’t get there by accident. They spent years imposing a standardized format, scrubbing every listing, and synchronizing rates and inventory in real time. They hold the cleanest hotel catalog in the world, and that’s the catalog the assistants now turn to.

They’ve also been the first to embed themselves in conversational interfaces. Several large chains are racing to do the same with their own data. When independent players all arrive at the same conclusion, it stops being a trend. It becomes a market law.

Most hotels are invisible to AI, and it’s not the AI’s fault

When a hotel shows up poorly in an AI response, the AI is rarely to blame. The problem is the absence of reliable, machine-readable information.

For a typical independent hotel, rates differ across the website, OTAs, and PMS. Room counts vary across systems. A pool that closed last spring is still listed somewhere. To a human, these are minor inconsistencies. To a machine trying to produce one reliable answer, they’re red flags.

Barely a quarter of AI responses about a hotel come from the hotel’s official content.

I’ve run the test repeatedly: same hotel, same question, two assistants, two completely different answers. The AI isn’t improvising. The underlying data simply isn’t structured or controlled. And the consequence is brutal: faced with conflicting sources, the agent discards the unreliable one, usually the hotel, and trusts the consistent one, which is almost always the OTA.

Every discrepancy across your channels is a free space left to a distributor or a competitor.

Agents need structure, not stories

Conversational AI doesn’t return lists. It offers two or three options, sometimes just one, with a justification. And the queries aren’t simple anymore.

Behind something like “boutique hotel near Saint-Germain-des-Prés with a spa,” a capable agent silently adds price, room specs, amenities, view, and a dozen other filters. Search intent has become multidimensional and heavily qualified. Only hotels with detailed, structured, current data can satisfy all the criteria, and only those hotels make the shortlist.

The advantage no longer goes to the hotel with the prettiest website. It goes to the hotel with the best-structured knowledge base. And the effect compounds: the more reliable a source is, the more often it’s cited, and the more it becomes the default.

A knowledge base is not a copy-paste of your website

Many hoteliers assume they can just “pull together what we already have,” the website, a couple of brochures, and hand it to the AI. It doesn’t work.

These sources are incomplete by design. Combined, they amount to maybe two or three hundred actual data points. A modern hotel is defined by roughly 4.000: accessibility, room-by-room amenities, dining, services, policies, and more. The gap is enormous. It’s the difference between ticking three boxes out of four and ticking four out of four.

And piling heterogeneous sources into a system that “reads everything” doesn’t add clarity; it adds confusion. The machine finds duplicates, contradictions, and outdated entries. It hesitates, then defaults to the intermediary. Keeping a patchwork like that current is, in practice, impossible.

Building a complete, standardized, exhaustive knowledge base and then maintaining it is unglamorous work. It won’t drive a campaign. It never really ends. But it’s non-negotiable. Trying to cut corners is like signing up for a triathlon and skipping the training. Structured data is swimming. Continuous updates are the cycling. Distribution is the running. In the age of AI, there are no shortcuts.

The cost of visibility is going up

The trajectory is clear. As discovery moves toward agentic search and intermediaries entrench their positions, hotels that AI can’t read will watch their organic direct traffic dry up, then have to buy it back.

Advertising is shifting from chasing clicks to buying visibility. For hoteliers, that means paying more every year to stay visible, bidding on their own brand name, and accepting commissions on bookings that, yesterday, would have come in directly.

Tomorrow, without a strong presence in agent responses, visibility will have to be purchased. The longer you wait, the higher the entry fee.

Data is the new distribution strategy

“Direct booking” no longer just means avoiding OTAs. It means being more visible, more understandable, and more actionable to AI agents than your competitors, and showing up in their recommendations.

Data quality, long treated as a back-office chore, has become the core of distribution. Hotels should start implementing the four CUPS steps now:

• Collect: a single source of structured data, a source of truth: a deep, detailed, standardized database, ready to be distributed.
• Update: static data quickly becomes stale; updates must be easy to manage and automatically propagate everywhere.
• Process: deliver structured data consumable by machines (and also humans).
• Share: distribute across every channel (OTAs, GDSs, web, bedbanks) via APIs and MCPs, and expose your hotel directly to the agents.

When a traveler asks an AI where to stay in your city, will your hotel be the trustworthy answer the machine recommends, or will it point to a dubious source it sets aside in favor of your distributor or your competitor?

Search has moved from links to answers. The intermediaries understood it first and got there first. The window is still open, but it’s closing as fast as the machines learn whom to trust. And that trust isn’t won with a prettier website. It’s won with clean structured data, and the unglamorous discipline of keeping it that way.

If you remember one thing, remember four letters: CUPS. Collect, Update, Process, Share.

It really is that simple.