AI Now Recommends Hotels, but Guests Still Book on the Hotel's Own Website
More than half of the luxury travelers surveyed by AGR booked a hotel they discovered through an AI assistant, yet most verify it on the hotel's own website and won't let AI pay. What this means for direct bookings, and what to do now.
BySebastián RondónCEO y fundador de KMS Agency Group
Two studies published this week, read side by side, paint a fairly clear picture of how a guest's journey now works. First, an Americas Great Resorts (AGR) survey of frequent luxury travelers found that 55.3% of those who answered the question had booked a hotel in the past year that they first discovered through an AI assistant. Second, Phocuswright research presented at its annual conference showed that 52% of U.S. travelers used AI to plan a trip, yet only 2% actually completed a booking inside an AI platform.
At KMS Agency, our takeaway is simple: AI is already deciding which hotels make the shortlist, but the booking still closes somewhere else. For an independent hotel, that "somewhere else" can be its own website, or it can be an OTA taking its commission. What tips the balance is how well prepared the hotel is.
What the data says
The AGR survey: AI discovers, the hotel website confirms
AGR sent its questionnaire to part of its audience of frequent luxury travelers and received 3,562 responses between November 2 and 19, 2026. It is a voluntary survey, not a representative sample of all travelers, and AGR says so itself. Even so, the patterns are highly useful for any hotelier:
- Discovery: 1,530 of 2,768 respondents (55.3%) say they booked a hotel they first learned about from an AI assistant.
- Verification: when AI recommends a hotel, the most common way to check it is the hotel's own website (59.8% of responses), followed by Google reviews (51.5%) and Tripadvisor (29.7%).
- Final decision: as the most trusted source for deciding, the hotel website again comes first (34.4%), ahead of Google reviews (31.2%). The AI assistant shows up in just 8%.
- Payment: 60% would not let AI book and pay on their behalf; another 25% would let AI hold the room but prefer to pay themselves.
- Errors: respondents report frequent AI mistakes, such as wrong prices (1,072 mentions) and incorrect amenities or restaurants (908).
The most-used planning tools were ChatGPT (53.6%), Google's AI Mode or AI Overviews (45.8%) and Gemini (25.2%).
Phocuswright: heavy use, little booking inside AI
According to TravelDailyNews's coverage of the opening of The Phocuswright Conference 2026, 95% of travelers who used AI say it influenced at least one trip decision, and 43% of those already familiar with these tools would consider booking directly inside an AI platform. Yet only 2% have done so (about 8% if Google's AI Mode is included). At the same event, United Airlines confirmed that it currently blocks AI agents from booking directly into its inventory: it supports AI-driven inspiration, but wants a human in the loop at the moment of purchase.
What this means for a hotel
1. Being on Google is no longer enough: you need to exist for AI
If more than half of this group of travelers booked a hotel that AI showed them first, a hotel that doesn't appear in those answers simply never makes the list. Assistants build their recommendations from public information: the website, reviews, listings and how consistent the data is. A hotel with incomplete, outdated or contradictory information has fewer chances of being recommended, and a higher risk of being described incorrectly.
2. The hotel website is where the booking is won (or lost)
For us, this is the most important finding: after the recommendation, travelers go to the hotel's website to check. That is where they decide whether to trust it. A slow site, with no clear rates, no current photos or no booking button that works on mobile sends that traveler straight back to the OTA, which closes the sale in two clicks.
3. AI errors are a reputation problem
When AI makes up a price or an amenity that doesn't exist, it isn't the hotel's fault, but it is the hotel that ends up welcoming the frustrated guest. The best defense is accurate, detailed information on your own channels: rates, policies, what each package includes and what it doesn't. The clearer the source, the less room AI has to fill in the gaps.
4. Travelers still want to talk to someone before paying
60% won't hand payment over to AI, and most want to confirm things on their own. That opens a huge opportunity for hotels that reply quickly and well on WhatsApp, chat or email, at any time of day. A traveler arriving from an AI assistant with doubts needs a concrete answer on availability, price and conditions, not a form that gets answered the next day.
What we recommend doing now
Based on these studies and on what we see in the hotels we work with, this is the order we suggest:
- Audit how AI describes you. Ask ChatGPT, Gemini and Google which hotels they recommend in your destination and how they describe yours. Note what is missing or wrong.
- Make your website the official source. Visible rates, clear policies, FAQs, what each package includes and a direct booking engine that works well on mobile.
- Look after your reviews and listings. Google comes right after the website as a source for verification and decision-making; replying to reviews and keeping your listing up to date is no longer optional.
- Reply in minutes, not hours. An AI agent on WhatsApp and on your website can handle availability, rates and common questions 24/7, and hand over to a team member the cases that call for judgment, such as exceptions or discounts.
- Give guests reasons to book direct. Clear benefits over the OTA (best rate, flexibility, a welcome touch) help turn verification into a direct booking.
- Track where each booking comes from. Asking guests how they found you and recording whether they came from an AI assistant gives you data that almost no hotel has today.
In another article, we explain how AI agents work in hotels and tourism and why human oversight remains key; it's the same logic we describe in our analysis of the 46% of tasks AI already handles on its own.
Conclusion
AI is becoming hotels' new shop window, but it isn't the cash register yet. Travelers discover with an assistant, verify on the hotel website and in reviews, and prefer to pay on their own. So the opportunity isn't in waiting for AI agents to book by themselves; it's in making sure the hotel shows up well in their answers and that, when travelers come to check, they find clear information and someone who replies right away.
Hotels that prepare their website, their data and their guest service for this journey today will capture more direct bookings. Those that don't will most likely watch them go through an OTA.
At KMS Hospitality, we help hotels organize their digital presence, their booking website and their guest communication with AI agents on WhatsApp, always with clear rules and with the hotel team in control. If you'd like to know how AI is presenting your hotel and what you can improve to win more direct bookings, let's talk.