Why Does AI Recommend Booking Sites Instead of My Hotel's Website?

In 5,334 AI answers about hotels that Drifter AI collected, 89.3% cited a booking or review site like Tripadvisor, Booking.com or Expedia, and 19.6% cited the hotel's own website. Booking sites describe every hotel in the same fields, with guest reviews on top, so an engine can compare them; your website is one source about one hotel.

That gap is wide, and it shows up on every engine we checked. It doesn't mean AI ignores hotel websites, though. Our data suggests the engines use different sources for different jobs, and the job of choosing between hotels mostly goes to pages built for comparison.

What did Drifter AI measure?

Drifter AI recorded which web pages AI engines cited when a traveler's question was about hotels but didn't name one. These are the questions a guest asks before choosing, like where to stay for a family beach trip or which resort has a good spa, with no hotel name in them. We ran 995 distinct questions, several times each, between May 10 and October 11, 2026, on Perplexity, OpenAI's models, Gemini, Claude and Google AI Overviews, for hotels that Drifter AI tracks.

A citation here is a source link the engine returned with its answer. We counted 62,179 of them across 5,334 answers. Another 5,791 answers came back without source links, so they aren't counted.

"Booking or review site" means the large booking, metasearch and review platforms: Booking.com, Expedia, Hotels.com, Tripadvisor, Trip.com, Agoda, Kayak, Trivago, Priceline and a few others. Drifter's wider source classifier also counts smaller online agents and tour sellers, and with those included the figure rises from 89.3% to 93.6% of answers.

Which booking sites does AI cite for hotels?

Tripadvisor was the booking or review site AI engines cited most, in 60.8% of the 5,334 hotel answers Drifter AI collected. Booking.com appeared in 44.5% and Expedia in 35.4%. Booking and review platforms made up 42.7% of all 62,179 citations. The hotel's own website made up 3.1%.

Tripadvisor is as much a review site as a booking site, so we took it out and counted again. Booking sites other than Tripadvisor still appeared in 80.4% of answers, so the picture holds without it.

It holds across engines too. The share of answers citing a booking or review platform ran from 72.0% on Gemini (1,006 answers) to 95.5% on Perplexity (2,266 answers). The share citing the hotel's own site ran from 11.9% on Claude (921 answers) to 27.0% on OpenAI's models (1,018 answers).

Why do AI engines lean on OTAs and review sites?

AI engines lean on booking and review sites because a question that doesn't name a hotel is a comparison, and those sites are built for comparison. A Booking.com or Tripadvisor page gives every hotel the same fields: where it is, room types, amenities, policies, a price range, a rating and hundreds of reviews written by guests. One results page lines up a dozen hotels. That's very close to what an engine needs to write "here are four good options for a family trip."

A hotel's own site is one source about one hotel, written by the hotel. The facts are often there, but spread across galleries, PDFs and a booking engine, and a page about one hotel can't line up the alternatives. I think engines treat it the way a careful friend would: useful for checking details once you've picked the place, less useful for deciding which place to pick.

The data fits that read. When the question named the hotel, its own site was cited in 57.1% of the 1,017 answers that mentioned it. When the question didn't name the hotel but the engine put it in the answer anyway, its own site was cited in 24.0% of those 3,856 answers. In 70.0% of them, the engine cited a booking or review site and never cited the hotel's site. The engine picked your hotel and gave the credit to someone else's page about you.

Hotel websites aren't shut out. Websites of other hotels and hotel groups showed up in 29.3% of the 5,334 answers, and 82% of those citations went to large international hotel groups, whose sites describe every property in the same layout. That looks like the booking sites' advantage at a smaller scale. We covered the experience side of this, where a hotel can say things no listing can, in Can a Hotel Become the Source AI Trusts?.

Which pages on a hotel's own site get cited?

When AI engines cited a hotel's own site, they cited the main property page most. Across 3,394 own-site citations, counting questions that named the hotel and questions that didn't, the main hotel page took 43%. Pages on a parent brand's site that list its hotels in an area took 13%. FAQ, location and about pages took 11%, restaurant and menu pages 10%, and spa, pool and beach pages 9%. Offers pages took 4%, and room pages under 2%.

Treat this as a hint, not a ranking. Most of those citations come from one hotel group's properties, so that group's site structure shapes the result, and the page types are our own grouping by URL. Still, the cited pages have something in common. They state a fact in plain text: what the hotel is, how to get there, what's on the menu, when the spa opens.

What can't this data tell you?

Drifter AI's citation data shows which pages AI engines cited, not which links travelers clicked, where they booked, or why an engine chose one source over another. A booking site in the sources doesn't mean the booking went through it.

Most of the answers are about resort hotels in one country. Answers about hotels elsewhere point the same way: across 297 Perplexity answers, a booking or review site was cited in 87.5% and the hotel's own site in 9.8%.

And 5,334 answers from 995 questions means the same questions repeat. It's a consistent pattern across every engine we checked, not a measure of all hotel search.

What should a hotel do about it?

A hotel can't make an engine cite it, but it can make sure the pages engines read about it, its own and the booking sites', state the facts a comparison needs. Here's where I'd start, based on what got cited.

1. Put the comparison facts on your own pages as text. Location, with real distances to the beach, the old town or the airport. Room types with size, beds and how many people they sleep. Amenities with hours and seasons. Policies on check-in, cancellation, children, pets and parking. And who the hotel suits, in a sentence a guest would recognise. If any of that lives in an image or a PDF, put it on the page. That goes for room pages too, which in our data were rarely cited.

2. Write the FAQ and location pages you'd want quoted. In our data they were cited far more often than room pages. Answer the questions your front desk gets every week, one plain answer per question.

3. Keep your Tripadvisor, Booking.com and Expedia listings accurate. In these answers, those listings were the source far more often than your website. Check that the description, room types, amenities and policies match what's true today, and fix what doesn't.

4. Run a check this week. Ask Perplexity, ChatGPT and Gemini five questions a guest would ask before choosing a hotel like yours, without your name in them. Note which pages each answer cites, and open the one that describes your hotel. If it's wrong or thin, that's your first fix.

None of this guarantees a citation, and our data can't show that it causes one. It does mean that when an engine compares your hotel with the others, the facts it reads about you are right, wherever it reads them.