Borrowed trust.
AI lends hotels trust consumers haven't extended to travel. Where it comes from, what it's worth, and what hotels owe in return.
When ChatGPT, Perplexity, or another AI assistant names a hotel as the answer to a question, the property inherits something it did not build. A recommendation from a system the traveler trusts. The hotel did not earn that trust. The assistant did. The hotel is borrowing it.
The traveler is not exactly lending it either. Only 18% of consumers say they fully trust AI for travel bookings. They act on the recommendation anyway. Both ends of the loan are on terms set somewhere else.
Low trust, high action
Quantum Metric's 2026 AI Experience Benchmark, a survey of 1,500 consumers in the US and UK, places that 18% at the bottom of the trust-by-category list. Consumers trust AI more for choosing clothing, books, and music (67%), for picking a cable, streaming, or mobile plan (64%), and for selecting financial products (45%). Travel sits below all of them.
Yet AI-referred travel traffic grew approximately 111% year over year. AI-driven cart values in travel are running at nearly twice the level of traditional traffic. The category that consumers trust AI for the least is the one moving the highest-value AI-referred transactions.
Something else is doing the work. The traveler is not acting because they have decided AI is reliable for travel. They are acting because the assistant they were already talking to gave them an answer that carried enough authority to follow. The conviction was lent, not built.
Where the trust is assembled
The assistant's authority on a specific hotel does not come from the assistant itself. It is assembled.
A March 2026 test by hotelemarketer.com asked ChatGPT to recommend a family hotel on Manhattan's Upper West Side, then ran the same query through ChatGPT's Tripadvisor app, then its Booking.com app. All three conversations returned the same top recommendation. On Tripadvisor's own site, that hotel ranked fifteenth. On Booking.com it did not appear on the first page. The preference was already there when the conversation began — baked into ChatGPT's training data, drawn from a corpus the hotel had no direct hand in shaping.
The corpus is large and skewed. Booking.com indexes roughly 29 million properties. Tripadvisor holds more than a billion reviews. Google Business Profile supplies structured data, hours, and ratings. Travel blogs frame the narrative. Reddit threads and YouTube videos record what other guests said. The hotel's own website is in there too, but it is one voice among many, and rarely the loudest.
When an AI assistant recommends a property, it is summarizing what other people — guests, journalists, OTA copywriters, the hotel's neighbors online — have said about it. The recommendation arrives at the hotel's site shaped by content the hotel mostly did not write. That is what is being trusted. That is what is being lent.
The visitor arrives decisive
In early 2025, AI-referred visitors spent roughly twice as long on travel and retail sites as visitors from any other source — checking the recommendation, validating, second-guessing. A year later, the same traffic spends about 25% less time on site than traditional visitors. The behavior reversed. Visitors stopped double-checking.
They arrive decisive. They have already taken the assistant's word for it. They are not shopping the page; they are completing a decision the assistant already made for them. They commit faster than other traffic and they expect the page to confirm what they were told.
They are also less tolerant of friction. The same Quantum Metric data shows AI-referred visitors are twice as likely to abandon a cart as traditional traffic at the first sign of trouble, and 81% say they will not return to a brand they discovered through AI after one bad experience. The conviction that brought them through the door is also what walks them back out — quickly, when the page does not match the recommendation.
The booking is worth more
When the experience confirms what the assistant said, the booking is unusually large. The 2x cart value Quantum Metric reports for AI-referred travel has two causes that compound. The visitor came pre-convinced. Travel transactions are heavy by nature. AI's most decisive buyers, when they arrive, spend the most.
That payment depends entirely on what happens on the property's site. The website needs to confirm the message the assistant was trained on. Photos that match the descriptions. Room language the guest would recognize from the recommendation, not the internal codes the PMS uses. Rates that hold. A booking engine that completes the transaction without breaking the spell.
None of this is new. The size of the booking hanging on it is.
Retail is the early signal
Retail is moving on AI commerce faster than hospitality, with bigger volumes and shorter cycles. What retail learns first, hospitality learns second.
One signal is already visible. The same Quantum Metric report finds 52% of digital teams investing in AI for customer support automation. Only 20% of consumers want that. Meanwhile, 46% of consumers want AI applied to search support — the discovery and pre-purchase layer. Fewer than 20% of digital teams prioritize it.
The investment is going one way; the demand is going the other.
For hospitality, the parallel is clear. Hotel groups building service chatbots, in-stay assistants, and concierge automation are working on a layer consumers have not asked for. The discovery moment — where AI is actually reshaping behavior, delivering a pre-convinced, high-value visitor to the booking flow — is where the data points.
The trust is borrowed. The visitor is decisive. The booking is worth twice a normal one. The question of what to build to keep all three is just starting.
by Markus Busch, Editor/Publisher 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 →