Published 21 July 2026 · Data collected June to July 2026
The new shelf: which hotels exist to AI and which ones don’t
The Stiplo Barcelona Study measured 659 registered properties across 27,360 AI travel answers to see which hotels AI recommends, which hotels it overlooks, who gets the link and why.
The story: AI travel answers are already reshaping hotel discovery, and the advantage is not evenly distributed.
Full PDF, methodology, source tables and republishable charts.
SCAN COMPLETE · 516 SURFACED
ONE DOT = ONE REGISTERED PROPERTY · N = 659
A fifth of Barcelona's register is invisible to AI
143 of 659 Barcelona properties never appeared in any answer, in any engine, in any language.
Independents are 8 times more likely to be invisible than chain hotels
The gap is discovery, not sentiment. Once surfaced, independents are recommended just as warmly.
Being named is not the same as winning the booking
Even when the hotel is recommended, its own site is among the cited sources just 7.3% of the time.
registered hotels, hostals and guesthouses across Barcelona
AI answers across real user travel prompts with 20x repetitions to control variability
real consumer AI assistants tested
languages: English, Spanish, Catalan
The visual story
Who gets the click, who never gets named, and how specialists win anyway. Three charts from the study, republish-able with attribution.
AIs link to OTAs more than hotel sites
Being mentioned doesn't guarantee the hotel's website is linked in the AI answer: OTAs get the chance to capture the booking.
% of answers linking to at least one
AI invisibility is concentrated
Most properties surface at least once, but the invisible tail is large and concentrated in independent, lower-category accommodation.
% never surfaced, by segment
Hotels can still win their niche
Breaking into the top 10% of hotels is hard, for most hotels the realistic way to win in AI visibility is owning their specific niche
Visibility boost for matching queries vs the average question
Distinct properties surfaced, by prompt type
| Category | Share of AI answers linking to at least one (%) |
|---|---|
| Editorial “best hotels” guides | 76% |
| OTAs & metasearch | 40.7% |
| The hotel’s own or chain’s site | 24.7% |
| International media | 17.8% |
| Reddit / forums | 12.6% |
| Social media (e.g. YouTube, Facebook) | 11.5% |
| Category | Share never surfaced in any AI answer (%) |
|---|---|
| All hotels | 21.7% |
| Rating: 3-star-plus | 4.7% |
| Rating: Lower category segment | 42.4% |
| Ownership: Independents | 32.4% |
| Ownership: Chains | 4% |
| Category | Visibility boost or distinct properties surfaced |
|---|---|
| Backpacker trip queries (a central hostal) | 419× |
| Football weekend queries (a stadium-side hotel) | 98× |
| Cruise stopover queries (a marina hotel) | 26× |
| Family city break queries (a family-brand hotel) | 12× |
| Geography and situation prompts | 410 properties |
| Direct recommendation prompts | 192 properties |
What really drives AI visibility isn’t your fashionable GEO checklist
Search visibility still matters, but the study points to reputation mass, authority and answerable content as clearer commercial levers.
MEASURED RELATIONSHIP WITH AI VISIBILITY. ASSOCIATION, NOT CAUSATION.
| Factor | Raw link | Adjusted read |
|---|---|---|
| Google review volumeThe strongest single correlation | Raw link0.64 | Adjusted read~0.35 |
| Search Visibility IndexSearch and AI visibility are related, but not interchangeable. | Raw link0.62 | Adjusted read~0.31 |
| Domain authorityThird-party web footprint matters. | Raw link0.55–0.57 | Adjusted read~0.22 |
| Answer-bearing pages existCrawlable practical pages beat absent content. | Raw link0.22 | Adjusted read~0.17 |
| Schema, llms.txt, AI-blockersUseful hygiene, but not proven budget items in this dataset. | Raw linkLow | Adjusted readNo independent signal |
What hotels can do to boost their AI visibility
Ask for Google reviews, not just OTA reviews. Google review volume was five times more important for AI visibility than OTA reviews volume.
Make sure practical pages exist before making them pretty. Hotels with pages answering real guest questions were visible 87% of the time vs 72% without. Polish added almost nothing.
Deprioritise the fashionable AI checklist. llms.txt, FAQ schema, AI-crawler settings and content freshness showed no meaningful signal.
The full report ranks every factor we measured and collects the playbook into four jobs for a hotel, including the fashionable tactics that showed no signal
Leave your email and we’ll send the detailed PDF with the complete hotel AI strategy.
One dataset, several stories.
The same data can serve hospitality trade, AI search, travel commerce and Barcelona-local coverage.
Independent hotels risk disappearing from AI travel answers
Chains are 37.6% of the register but receive 62.2% of surfacings. Independents are most of the market, but only 37.8% of surfacings.
AI has new gatekeepers, and nobody audits them
Independent editorial guides and listicles are about 44% of citations, the largest single bucket, including lists with no identifiable publisher; official tourism boards are almost absent. Reddit is the single most-cited domain, Booking.com the most-present (14.3% of answers).
AI names hotels, then often hands the click to someone else
The commercial question is not only whether AI recommends a hotel. It is whether the booking path reaches the hotel or an intermediary.
The short answers, straight from the data
How many hotels are invisible to AI?
In Barcelona, 21.7% of the official register: 143 of 659 registered properties never appeared in any of 27,360 AI travel answers, in any engine or language.
Invisibility is concentrated: 4.7% of 3-star-plus hotels never surfaced, against 42.4% of the lower-category segment.
Do AI assistants favour chain hotels over independents?
On discovery, yes: independents are 8 times more likely to be invisible to AI than chains (32.4% vs 4.0% never surfaced). Chains are 37.6% of the register but receive 62.2% of surfacings.
The gap is discovery, not sentiment. Once surfaced, independents are recommended just as warmly.
Which sources do AI assistants link when recommending hotels?
Editorial guides dominate: an independent guide or listicle shows up in about two-thirds of answers, an OTA in roughly half, and a hotel-or-operator-owned page in about 40%.
Booking.com is the single most-present domain (14.3% of answers), but Reddit is the most-cited. Among the answers that name a hotel, 48.1% carry an OTA or aggregator source and only 29.6% any hotel's own domain; the named hotel's own site is among the sources just 7.3% of the time.
What actually makes a hotel visible to ChatGPT and other AI assistants?
Google review volume shows the strongest single correlation with AI visibility (0.64), ahead of the Search Visibility Index (0.62) and domain authority (0.55 to 0.57). Hotels with crawlable pages answering real guest questions were visible 87% of the time vs 72% without.
The fashionable checklist underperforms: schema markup, llms.txt and AI-crawler settings showed no meaningful signal in this dataset. Association, not causation.
Can a small hotel still win in AI search?
Yes, by owning a niche. Geography and situation prompts surfaced 410 distinct properties, more than double the 192 surfaced by direct recommendation prompts.
One central hostal was 419 times more likely to surface for backpacker queries than for the average question; a hotel beside the stadium 98 times more for football trips. The realistic play is owning your specific niche, not cracking the generic top 10.
How Stiplo measured the AI shelf
Enough detail for a journalist to assess credibility quickly, with full methodology and source tables available under embargo.
Visibility
means named in a recommendation or list context. It is not the same as endorsement or booking.
Matcher accuracy
is validated at 0.93 to 0.95 recall and 0.95 to 0.97 precision across held-out human-labelled checks.
Adjusted figures
strip out major prominence proxies such as star category, chain membership and domain authority.
One city, one vertical,
free consumer AI surfaces. Other markets and paid research modes may behave differently.
“According to the 2026 Stiplo Barcelona Study, 143 of Barcelona's 659 registered properties (21.7%) never appeared in any of 27,360 AI travel answers across ChatGPT, Perplexity, Google AI and Gemini.”
“The Stiplo Barcelona Study found independent properties are 8 times more likely to be invisible to AI assistants than chain hotels: 32.4% of independents never surfaced, against 4.0% of chains.”
“Across 27,360 AI travel answers analysed by Stiplo, a recommended hotel's own website was among the answer's cited sources just 7.3% of the time.”
The Stiplo Barcelona Study: How AI Decides Which Hotels Exist. Stiplo, 2026. https://stiplo.io/barcelona-study
Figures and charts are republishable with attribution and a link to this page.
“Hotels are used to checking where they rank in Google. The AI layer is a new shelf: it names some properties, forgets others, and often sends the click to someone else.”Carlo Del Mistro, Founder, Stiplo. Former Chief Digital Officer at Ennismore.
Request full data
Leave your email and we’ll send the detailed PDF: full methodology, source tables and the republishable chart pack.
Media contact
For methodology questions, property-level cuts or interviews.