Found, but not bookable: the gap between AI visibility and AI bookings for independent hotels

In 2026, independent European hotels face a question that did not exist a year ago: if a traveller asks an AI assistant where to stay, does the property show up? And if it does, can it be booked right there, without going through someone else? Three news items from recent weeks, read together, show that the answer can be no at three different levels, and that the three levels call for different actions.

The frame comes from Skift, reporting from its Data + AI Summit in London on 6 October: AI is changing how travel is discovered and organised, but the volume of bookings it generates is still marginal. According to the report, Booking Holdings states that fewer than 1% of room nights come from AI channels, Bernstein estimates around 1.6% for Marriott, and a Skift/McKinsey study indicates that AI use drops to about 6% at the booking and payment stage. Fifty-one per cent of travellers, according to a survey cited by Skyscanner's CEO, say they do not trust the reliability of AI recommendations. Before moving any budget, then, it is worth remembering the proportions: today we are under 2%.

That does not make the topic any less important, because discovery is the moment when it is decided who makes the list.

First level: being found

The most useful study is the one by Stiplo on Barcelona hotels, published on 21 July 2026. It queried four free AI assistants (ChatGPT, Google AI, Gemini and Perplexity), in English, Spanish and Catalan, about 659 properties, for a total of 27,360 responses.

The main numbers:

• 143 properties out of 659 (21.7%) never appeared in any response.
• Among independents, the share of properties never mentioned is 32.4%, against 4.0% for chains: roughly eight times as much.
• Chains make up 37.6% of the sample, but receive 62.2% of the citations.
• When an independent does appear, it is recommended with the same conviction as chains: the problem, according to the study, is discovery, not bias.
• The variable most associated with visibility is the number of Google reviews (raw correlation of 0.64, about 0.35 after adjusting for stars, chain affiliation and domain authority).
• Editorial guides of the "best hotels in…" type appear among the links in about three responses out of four (76% in the study's chart, "about two thirds" in the FAQ on the same page: the source is not consistent). The hotel's own website is cited in 7.3% of the responses that mention a property.
• Pages that answer concrete guest questions have higher visibility (87% versus 72%). Schema markup, llms.txt files, AI crawler settings and content freshness showed no meaningful signal.

Caveats: Stiplo sells audits and optimisation services to hotels, and its recommendations coincide with what it sells. The study covers one city and free tools, and it measures associations, not causes. The full report sits behind a form. As far as the research I carried out shows, there is no equivalent data for Italian cities: the logic is plausible, but it remains a hypothesis to be tested locally.

Second level: being bookable

Since 27 August, in the United States, Google lets users search for and book hotels directly in AI Mode, paying with Google Pay. There are ten launch partners: five OTAs (Booking.com, Expedia, Hotels.com, Priceline, Trip.com) and five hotel groups (Choice, Hilton, IHG, Marriott, Wyndham). Google is not the merchant of record: billing and customer service remain with the hotel or the platform.

Independents are not among the partners. A commentary on Hotel Online sums it up as a gap between "visibility" and "bookability": a property can appear in the responses, but the traveller is redirected to a partner's checkout. The author, who describes himself as an AI consultant for hotels, is interpreting the partner list: Google has made no statements about its plans for independents. The launch covers only the United States, and the sources give no timing for other markets.

Third level: being described well

In China the picture is already more advanced. According to Hospitality.today, the AI agent of Fliggy (Alibaba's travel platform) has a dedicated access point for Marriott, through which it can search by brand, destination or name and also sell packages such as afternoon tea and spa, not just rooms. All other hotels, chains and independents alike, go through a single search that filters only by star category, bed type, price, property type and distance from an attraction: five fields. IHG, Hilton and Hyatt have stores on Fliggy, but there is no dedicated command for the agent.

Fliggy reports an increase in AI bookings of more than 800% over the holiday period, driven by attraction tickets, without providing any hotel data. These are company figures, reported by a niche trade outlet.

A reading: three levels, three actions

The link between the three stories is my own reading, not the sources', which do not cite each other. It seems useful, though, because it separates different problems:

1. Discovery. It depends on reviews, on mentions in guides and forums, on pages that answer specific questions. Here the action is within an independent's reach.
2. Bookability. It depends on agreements and integrations that a single hotel does not control. Here the action is indirect: choosing technology partners (channel manager, booking engine, OTAs) that will be included in the new channels, and keeping the direct booking path fast and competitive.
3. Description. It depends on how the property's data is structured: if a platform describes a hotel with five fields, its character never reaches the customer.

What a property can do today

• Test. Ask AI assistants, in Italian and in the languages of your own markets, the questions a guest would ask ("where to stay in Piedmont with children and no car") and note whether and how the property appears.
• Work on Google reviews: volume, replies, variety of contexts.
• Create pages that answer concrete questions (how to get there without a car, what to do when it rains, who the property suits), rather than generic pages. The study suggests that situational queries, such as a sports weekend, open more doors than generic prompts.
• Do not invest in technical files for AI crawlers on the basis of this study. Stiplo found no effect from llms.txt and schema markup; it is a single study, but at least there is no evidence in their favour.
• Keep prices and information consistent across all channels and, when you don't, state the reason.
• Do not move budget in proportion to the hype: with less than 2% of bookings coming from AI, measurement is what counts.

What we don't know

There is no Italian data on visibility in AI assistants. We do not know if and when bookings in Google AI Mode will arrive in Europe. Fliggy's numbers are company-reported. And every study from a hotel service provider, this one included, has to be read bearing in mind what the provider sells.

To go further on Officina Turistica: Meta Muse e gli agenti AI per i viaggi and Booking, Fogel e la fiducia.

FAQ

Are independent hotels really less visible to AI than chains? According to Stiplo's study of 659 Barcelona properties, yes: 32.4% of independents never appeared in the responses, against 4.0% of chains, roughly eight times as much. When they do appear, they are recommended with the same conviction. The study covers one city and free tools, and comes from a hotel service provider.

Can you already book a hotel inside an AI assistant? In the United States, yes, since 27 August 2026 in Google AI Mode, with ten partners: five OTAs and five hotel groups. Independents are not among the launch partners. No dates are known for arrival in Europe.

How much do AI channels weigh on hotel bookings today? Little: Booking Holdings states fewer than 1% of room nights come from AI channels and Bernstein estimates around 1.6% for Marriott. The weight is greater at the discovery stage than at the booking stage, but the data is still partial.

*cover image generated with AI