The hospitality industry is entering a new phase of search — one where the first result is not a list of links but a direct recommendation from an AI. We wanted to understand how that recommendation engine works, who it favours, and what individual properties can do about it.
So we ran 450 queries. Here is what we found.
Methodology
We tested four AI models: **ChatGPT (GPT-4o)**, **Google Gemini**, **Perplexity**, and **Mistral Large**. Across four EU countries and Switzerland: Croatia, Slovenia, Austria, Germany, and Switzerland.
Each query was structured as a natural traveller request: *"Where should I stay in Split in August?"* or *"Best boutique hotel in Ljubljana under €150."*
**Sample size:** 1,337 accommodation websites audited. 450 queries executed.
The Core Finding
**94.3% of accommodation websites receive zero AI recommendations** in our query set.
That number is not a rounding error. It means that for every 18 properties in a destination, only one is mentioned — and that one is almost always an OTA listing, not the property's own website.
What AI Models Actually Use
After correlating recommendations with website attributes, we identified the strongest predictors of AI visibility:
- **Schema.org markup** (Hotel, LodgingBusiness) — 6.2× lift
- **English-language content** — 4.1× lift
- **Direct booking engine presence** — 3.3× lift
- **Review aggregate score > 4.5** — 2.8× lift
- **Structured FAQ content** — 2.1× lift
- **Core Web Vitals pass** — 1.9× lift
The OTA Problem
OTAs are winning because they have all six attributes — at scale, for every property they list. Booking.com has schema markup, English content, direct booking infrastructure, review aggregation, FAQ content, and fast-loading pages.
Individual properties mostly have none of these.
The Market Breakdown
| Country | AI Visibility Score | OTA Dependency | Direct Booking Score |
|---|---|---|---|
| Austria | 40.1 | 22.4% | 14.2% |
| Germany | 38.9 | 24.1% | 12.8% |
| Switzerland | 37.6 | 25.1% | 11.9% |
| Slovenia | 36.2 | 26.3% | 10.7% |
| Croatia | 34.8 | 30.7% | 9.2% |
What Works
The properties that do appear in AI recommendations share a common profile: - Schema markup implemented - Booking engine embedded on own domain - Content written in the language of the query - Review score visible in structured data - Load time under 2 seconds
None of these are expensive to implement. Most are free. The gap between the invisible 94.3% and the recommended 5.7% is largely a technical literacy gap, not a resource gap.