Barcelona Study: Independent Hotels 8x More Likely Invisible to AI

Independent hotels in Barcelona were eight times more likely than chains to be completely unknown to AI assistants, according to a study of 27,360 AI travel answers presented at Skift's Data + AI Summit in London this month.

Carlo Del Mistro and the team at Stiplo ran the most rigorous test yet of how AI recommends hotels: 659 Barcelona accommodations, four AI assistants, three languages, and every question asked 20 times, because AI answers change between runs.

The core finding overturns one common assumption. Once an AI assistant recognized an independent hotel, it recommended it about as often as a chain. The barrier is not product or brand quality. The AI has simply never heard of the property.

What did the Barcelona data show?

The numbers quantify a visibility gap that produces no traceable lost booking:

• About 1 in 5 hotels was never recommended at all, in any language, by any AI.
• About 20% of hotels captured more than 80% of all AI mentions.
• About 68% of what AI said about hotels came from guides and listicles; only about 7% came from hotel websites.
• Only about 7% of AI recommendations linked to the hotel's own site. Most pointed to OTAs — Booking.com most often — plus TripAdvisor and listicles.

For operators, the direct-booking engines they have invested in for years supply roughly 7% of what AI tells guests about their properties.

Is AI already booking rooms?

Google's AI Mode now lets US travelers book hotels without leaving the conversation. Its launch partners are the large chains and the OTAs, not independents.

Meta's Muse, the assistant built into WhatsApp, Instagram and Facebook, is rolling out booking as well. It completes hotel bookings through Expedia and Hotels.com rather than through hotel systems. When Skift tested Muse on the Hilton, Marriott and Hyatt websites, the assistant hit errors before it could reach a rate.

The scale is still small. Booking Holdings says less than 1% of its room nights come from AI; Bernstein estimates about 1.6% at Marriott. But about a third of AI hotel recommendations come from what the model already "remembers" from training data — and that memory is forming now, while volumes remain low.

The study's warning for independents is direct: hotels that ignored Google in 2008 and OTAs in 2012 spent the next decade buying back visibility at 15 to 25% commission. The properties that enter AI's memory in 2026 will be the defaults in 2028.

Does "AI optimization" tooling work?

The Barcelona data found no measurable independent effect from the products hotels are being sold — LLMs.txt files, crawler settings, FAQ markup and schema packages.

What did move AI visibility:

• Google review volume, the strongest single driver.
• Backlinks and existing Google rankings.
• Placement in the guides and listicles the assistants actually read.
• Clear website content answering real guest questions.

What should operators do now?

The study sets a sequence. Test the hotel in ChatGPT, Gemini, Grok, Perplexity, Meta AI and Google's AI Mode with real guest questions, asked several times, to identify the actual competitive set. Make review volume a weekly KPI, built into checkout and post-stay messaging. Pitch the local listicles the assistants keep citing. Cover parking, accessibility, quiet rooms, distances and late arrival in plain language on the site. Only then add AI booking and guest-messaging tools.

The open question the report leaves with management teams: when a guest asks AI where to stay in your city, is your hotel in the answer — and if nobody knows, that is the answer. Del Mistro's team is now recruiting independent and boutique properties of 10 to 250 rooms for live AI-readiness diagnoses, with limited places.