The search begins with operational truth
Over the last few years, choosing a hotel stopped being a small part of travel for me and became something much closer to a permanent operational task. I lived in Batumi and Tbilisi, then in Goa and Mumbai, then kept moving between countries, islands, and cities: Phuket, Krabi, Samui, Phangan, Ho Chi Minh City, Vung Tau. At some point I realized that the phrase "digital nomad," for all its overuse, described my reality quite well: a laptop, a suitcase, constant movement, work done from hotel rooms, different countries, different rules, different payment habits, and very different ideas of what a "comfortable hotel" actually means. To be honest, I did not exactly choose this life; in many ways, it chose me. But I am happy with it. Over the years I have seen dozens of strange, beautiful, uncomfortable, cozy, chaotic, and completely unexpected places that an ordinary tourist might never notice.
My loyal friend Richie, a pug, travels with me, and that is a separate chapter in the story. He is wonderful, I love him, but he also makes life much more complicated: flights, export documents, country rules, airline rules, veterinary papers, weight and breed restrictions, and endless confirmations before check-in. Finding a hotel that will calmly accept a stubborn ten-kilogram pug, allow a late-night arrival, avoid a surprise deposit, provide decent internet, and not turn out to be a noisy box by the road is no longer a simple accommodation search. It becomes a small investigation. That is where you quickly learn that nice-looking filters like "pet-friendly," "free Wi-Fi," and "late check-in" often hide more uncertainty than they resolve.
For an ordinary tourist, a hotel often starts with mood: the view, the pool, the breakfast, the neighborhood, the photos, the reviews, the price. For someone who lives between places, a hotel starts with much more practical questions. Can I actually check in after midnight if the flight is delayed? Will they accept the dog not "in theory," but at the front desk when I am standing there with a suitcase and a tired pug? Is there a real table in the room where I can work, or just a narrow decorative shelf under a mirror? Is the internet really good enough for calls and file uploads, or is it only a Wi-Fi icon in the amenity list? What is the deposit? Which payment methods work? Is it noisy? Can I sleep properly? Will the "quiet room" turn out to be above the restaurant or next to construction? These are not luxuries, and they are not exotic demands. They are ordinary conditions that decide whether a specific hotel will work for a specific person.
Filters create the illusion of clarity
The most frustrating part of the old hotel search was not that there was too little information. There was too much. Booking, Google Maps, reviews, OTAs, the hotel website, photos, old articles, other people's lists, short WhatsApp replies — all of this creates the feeling of choice, but not always the feeling of clarity. Reviews are often outdated, emotional, purchased, or written by people with completely different expectations. One guest gives five stars because they liked the breakfast, another complains about noise, a third writes that the Wi-Fi was "fine," but there is no way to know what "fine" meant for them. A good newly renovated hotel may have almost no review history and look weaker online than an older, average property with thousands of ratings. The hotel website may tell a beautiful story but avoid exact rules. The OTA may say one thing, Google another, the message thread a third, and the front desk something else entirely.
That was when I started using ChatGPT more often, not as a toy and not as a replacement for Booking, but as a way to narrow the chaos. I was no longer writing "find me a good hotel." I was giving it my real conditions: the city, the area, a dog of about ten kilograms, late arrival, decent internet, a place to work, and preferably no surprises with payment or deposit. The more precise the request became, the clearer it was that the logic of choice itself was changing. I was not looking for the "best hotel" in the city anymore. I was looking for a hotel that fit my exact situation. That is a completely different task, because "best" in a general sense can be useless if the hotel does not accept dogs, does not allow late check-in, does not have a proper desk, or cannot clearly explain payment rules.
But then another problem appeared. ChatGPT often helped narrow the search, but the more complex the scenario became, the more often the model moved toward large chain hotels, obvious options, or properties with more structured and safer signals online. Small good hotels that might have been perfect often failed to pass through this confidence filter. Not because they were worse. Not because they had poor service. But because their reality was poorly expressed in data. In one case, there was no precise pet policy. In another, late check-in was listed as "upon request," but without a clear process. Somewhere else, Wi-Fi was mentioned, but with no evidence that it was suitable for work. In another case, the rules on the website and the OTA looked different. For a human, that is a reason to clarify. For a model, it is a reason not to take responsibility.
Travelers increasingly choose scenarios, not categories
This was where I first really saw what later became the foundation of Evidentity: travelers increasingly choose scenarios, not categories. In the past, the human did the work. They read, compared, messaged, clarified, called, took risks, made mistakes, arrived, and discovered the real situation on site. AI simply takes that work onto itself and performs it faster, more strictly, and with far less tolerance for uncertainty. A person may still tell themselves, "Fine, I'll risk it." A model is more likely to choose the safer option. A person can message the hotel. In most cases, the model simply substitutes another property. A person can guess that "pet-friendly" probably means "small dogs are allowed." A model should not build a recommendation on that guess.
This is the scenario economy in hospitality. It does not begin with AI, but AI makes it visible and commercially unforgiving. The request "find me a good boutique hotel" creates one competitive field. The request "find me a quiet hotel where I can work, arrive after midnight, and check in with a dog" creates another one entirely. In that second field, not every hotel in the city participates. Only the hotels whose facts allow AI to satisfy those constraints with confidence remain in the choice set. A hotel may be excellent, but if its scenario signals are not clearly expressed, it can disappear from the decision before the guest ever reaches the website, Booking page, or price comparison.
The loss is invisible in analytics
For a hotel owner, this is especially unpleasant because the loss is almost invisible. If a person visits the website and does not book, analytics can show something. If a person opens the OTA page and leaves, that is at least part of the familiar funnel. But if AI never includes the hotel in the answer because it cannot confirm the required conditions, ordinary analytics show nothing. There is no click, no bounce, no abandoned form, no direct signal. The booking still happens. It simply happens at another hotel. Demand does not disappear; it is routed toward the place where the model feels more confident.
That is why the old idea of "we have a website, so AI will understand us" feels weaker every month. The website remains important: it sells atmosphere, trust, brand, visual feeling, and direct booking. But for AI, it is not enough. The model needs to understand operational reality: what the hotel supports, what it does not support, which limitations exist, which scenarios it can actually satisfy, where stable facts end, and where real-time commercial data such as prices or room availability begins. Beautiful text can persuade a person, but it does not always give a machine the right basis to recommend with confidence.
The Pug test
I think about this through a simple "Pug test." If AI cannot confidently understand whether someone can arrive late in the evening with a ten-kilogram dog, work properly from the room, avoid a hidden deposit, and reach the official booking path without confusion, then the hotel is not yet ready for the new logic of recommendations. That does not mean the hotel is bad. It may be the perfect fit. But its reality is not expressed in a way that a machine can safely use in an answer. For a human, that is an inconvenience. For AI, it is a reason to choose someone else. Evidentity grew out of this gap between how hotels describe themselves online and how real travelers make decisions. We build a governed AI Profile — a managed layer of operational truth for a hotel that translates the real conditions of the business into a form recommendation systems can understand. The goal is not to make a hotel "sound better" to AI. The goal is to make its facts, rules, scenarios, and boundaries clear enough for the model to understand when the hotel truly fits, and to route the guest toward the official booking path instead of a clearer competitor.