How ChatGPT picks hotels: the 810-query study
If you ask ChatGPT for a hotel for your next trip, it won't look at your website first. It will reconstruct an opinion from what others say about you: booking platforms, blogs, forums and reviews. That is exactly what the first empirical study on the subject measured, and its numbers should reshuffle the priorities of any accommodation business.
The study: 810 queries, 3 assistants, 6 destinations
The report "The Signals Behind Hotel AI Recommendations", published by Cloudbeds in 2025 and updated in 2026, ran 810 queries against ChatGPT, Gemini and Perplexity — the three platforms concentrating around 98% of AI-originated web traffic. It covered 6 destinations (Barcelona among them), with three prompt types per destination and every query repeated 15 times per platform. The result: a set of 145 hotels that showed up consistently in the answers.
Worth saying upfront: Cloudbeds is a hotel software vendor, so it has skin in the game. But the methodology is published, and today it is the study cited by the industry's leading media.
Where the AI gets its information
The breakdown of citations used by assistants when recommending hotels came out as follows:
| Source | Share of citations |
|---|---|
| OTAs (Booking, Expedia...) | 55.3% |
| Travel blogs | 19.2% |
| Official hotel websites | 13.6% |
| Forums and communities (Reddit, Tripadvisor) | 6% |
| Video reviews | 1.6% |
Two readings that usually get missed. First: that 13.6% share for the hotel's own website is actually relatively good news, because it beats the cross-industry average of 9% for brand-website citations. In hotels, your official site matters more than in other industries. Second: more than 40% of brand impressions on ChatGPT and Perplexity come from third-party citations, according to Mathias Ptacek, founder of Rankscale, the tool used in the study. Translated: what is said about you outside your website weighs as much as your website — or more.
Every assistant has its own character
Another under-reported finding is that the three platforms don't drink from the same wells. Gemini clearly favours Booking.com and is the only one citing Reddit and YouTube. ChatGPT references official hotel websites the most — the friendliest one to the direct channel. Perplexity mostly cites Tripadvisor, with which it holds a commercial agreement.
This has a practical consequence: there is no single "AI ranking strategy". If your market leans on Gemini, your Booking listing and your YouTube presence matter more; if it leans on ChatGPT, your official website has more headroom. An Ahrefs analysis of 78.6 million AI searches points the same way: YouTube and Reddit are the second and third most-cited domains by assistants.
What the recommended hotels had in common
Clear patterns emerge from the profile of the study's 145 properties. 72.4% belonged to a brand or group, with an average visibility advantage of 4.43 percentage points — but independents still held more than 25% of recommendations. All had a high volume of reviews, with an average sentiment of 75 out of 100. And their digital footprint was broad: 98% were present on YouTube, 97% on travel blogs and 95% on Reddit, even when not directly cited.
The underlying message: reputation and distributed presence are not a complement to your AI visibility strategy. They are the strategy.
Four things you can do this week
- Ask the assistants yourself. Open ChatGPT, Gemini and Perplexity and run the questions your customer would ask ("family hotel in [your city] near the beach", "dog-friendly hotel in [your area]"). Note whether you appear, on which platform, and which sources are cited. It is the cheapest audit there is.
- Look after recent reputation. The recommended hotels had many good reviews. A steady flow of fresh reviews weighs more than a brilliant but stalled history.
- Work on third-party sources. Travel blogs, local media, recommendation lists, YouTube, Reddit: that is where the AI gets almost half of what it says about you. A mention in a destination guide is worth more than a new paragraph on your homepage.
- Make your website quotable. Concrete, verifiable facts (opening times, policies, services, indicative prices), structured data and content that answers real questions. I apply this same principle on my own website — from JSON-LD markup to an llms.txt file — and it is the one part of the system you fully control.
The uncomfortable conclusion
For twenty years the question was "how do I show up on Google?", and the answer depended mostly on your website. The new question is "what does the AI say about me?", and the answer depends on a whole ecosystem: your OTA listings, your reviews, what blogs write, what gets discussed on Reddit. You can work on it, but you cannot control it from a single place. Whoever grasps that board change first will play the next few years with an advantage.