How Does AI Recommend Hotels? We Tested 450 Queries Across 4 Models to Find Out

The way travelers find and book accommodation is changing. For the past two decades, the journey followed a predictable pattern: type a keyword into Google, scan a list of blue links, visit a few websites, compare prices, book. That journey is being replaced by something fundamentally different.

When a traveler asks ChatGPT, Perplexity, or Google's Gemini to recommend a hotel in Vienna, a cabin in the Black Forest, or a beach resort in Croatia, the AI doesn't return a list of ten links. It recommends three to five specific properties by name. Either your property is on that list, or it isn't.

There are no blue links. No page-two rankings. No "above the fold" positions. AI search is binary: recommended or invisible.

This isn't a future scenario. Research shows that information-seeking queries account for nearly 50% of all ChatGPT conversations, with direct question-asking behavior representing 37% of all interactions. AI-powered travel search queries are, on average, twice as long as traditional Google keywords — conversational, specific, and intent-rich.

To understand what determines whether AI recommends your property, and whether your website is ready for this shift, we conducted the largest AI visibility study in European hospitality.

About the Research

Between August 2025 and February 2026, we conducted two original studies:

Study 1 - AI Response Analysis:

450 hospitality prompts tested across four AI models (GPT-5, Gemini 2.5, Perplexity, and Mistral), generating 3,600 AI response records containing 20,370 URL citations and 13,859 brand mentions.

Study 2 - Website Audit:

1,337 accommodation websites audited across 30,935 individual pages, evaluating content quality, technical SEO, schema markup, booking funnels, and trust signals.

Five countries: Germany, Austria, Switzerland, Slovenia, and Croatia. Three query languages: English, German, and Croatian. Ten travel intent categories, from luxury and business travel to budget city breaks and adventure tourism.

The full dataset, methodology, and statistical analysis are in the report. This article covers the key findings and what they mean for accommodation operators, hotel marketers, DMOs, and hospitality technology providers.

Who Owns AI Travel Search?

When a traveler asks AI to recommend accommodation, who gets cited?

The answer is clear: OTAs dominate, independents survive at the margins, and one platform towers over everything else.

Online Travel Agencies receive 22.9% of all AI-cited URLs — the single largest classified category. Add STR platforms like Airbnb and VRBO, and the total intermediary share reaches 30.3%.

Independent hotel websites receive 11.8% of citations. Hotel chain websites receive just 4.3% — less than one-third of independents. Review and UGC platforms (primarily TripAdvisor) account for 5.8%. DMOs and tourism boards contribute 3.9%.

For every one independent hotel URL an AI model cites, it cites roughly two OTA URLs.

The concentration is extreme. Just 119 domains account for 50% of all AI travel citations. 2,981 long-tail domains appear only once or twice in the entire dataset. Only 86 out of 4,043 total domains were cited across all five countries — the truly "AI-canonized" hospitality brands.

Booking.com's Dominance in AI Search

No single finding in this dataset is more striking than Booking.com's position in AI-powered hotel search.

2,962 total citations. 95.3% prompt coverage. 14.5% of all URLs.

Booking.com appeared in 95.3% of all 450 queries we tested. It was absent from only 21 queries — and those gaps reveal specific niches where OTA inventory doesn't match: converted monasteries, ultra-luxury boutique properties, niche eco-lodges with strong editorial authority.

The practical implication for AI search optimization in hospitality is straightforward: for any property not on Booking.com, AI models have near-zero probability of citing them. A complete, up-to-date Booking.com listing is the single highest-impact AI visibility action any accommodation operator can take.

But it also means something harder: if the only place AI can find your property is Booking.com, the citation — and the commission — goes to them, not to you.

How Different AI Models Recommend Hotels

Each AI model exhibits a distinct citation personality. These differences are statistically significant and have direct implications for hotel AI search optimization strategy.

GPT-5 (ChatGPT) produces the longest responses (439 words average), cites the most URLs per response (10.6), and offers the best direct booking outcomes (20.6% direct booking score). However, it's the least consistent model — results vary significantly between identical queries.

Gemini 2.5 (Google) has the highest OTA dependency of any model at 29.4%. This is especially significant because Gemini is integrated into Google Search, making it the AI model most travelers encounter first. For independent hotels and direct booking strategies, Gemini is the hardest battleground.

Perplexity is the most predictable model (Jaccard similarity: 0.47). Properties that break into Perplexity's citation set can rely on repeat visibility. It has the lowest OTA reliance (20.5%) and the highest review/UGC citation share (17%), making TripAdvisor presence especially valuable for Perplexity visibility.

Mistral produces zero URL citations but still generates 2.5 brand mentions per response, proving that brand recall in AI operates independently of website indexing.

You can't optimize for "AI search" as a single channel. Each model has different behaviors, different source preferences, and different outcomes for direct versus OTA booking.

How Query Language Affects AI Hotel Recommendations

Every query in this study was asked in three languages — English, German, and Croatian — for the same country and intent. This creates a controlled experiment: does the language of the question change what AI recommends?

The answer is yes, significantly.

OTA dependency for Austrian hotels by query language:

• English: 11.4%
• German: 10.5%
• Croatian: 20.6%

That's a 10 percentage point gap between German and Croatian — for the exact same country, same hotels, same AI model. This pattern holds across all five markets. Croatian-language queries consistently produce the highest OTA dependency, even for non-Croatian destinations.

The hypothesis: Croatian-language training data for DACH markets is less rich in local provider content. When AI models have less local content to draw from in a given language, they fall back to globally indexed OTA pages.

For hotel AI search optimization, the practical implication is clear: if your website exists only in one language, you're invisible to travelers searching in other languages. English and German versions of your site capture the "direct-friendly" language channels where AI is more likely to cite independent sources.

What Travelers Ask Determines Who AI Recommends

The intent behind a travel query dramatically shapes which providers AI models cite. This is the most actionable finding for hotel content strategy and AI search optimization.

Best intents for direct hotel bookings in AI search:

• Business travel: 28.8% direct booking score
• Luxury / Romantic: 22.8%
• Wellness: 22.3%

Worst intents for direct booking:

• Event / Festival: 6.1% direct score (39.4% OTA dependency — the highest)
• STR / Rental: 7.2% (platform-dominated by Airbnb and VRBO)
• Budget City: 9.4%

For hotels serving business, luxury, or wellness travelers, a well-crafted direct website with authoritative content genuinely competes with OTAs in AI search. For event-adjacent properties, AI defaults almost entirely to OTA inventory — making event-specific landing pages with direct booking CTAs essential.

AI also mirrors price sensitivity by intent. Pricing information appears in 100% of wellness queries, 83% of nightlife queries, and 76% of STR queries — but only 36% of unique-stays queries. If your wellness property doesn't display pricing on your website, you're invisible in the category where AI always talks about price.

94.3% of Hotel Websites Are Invisible to AI

The second phase of this research audited 1,337 accommodation websites across 30,935 pages. The findings paint a stark picture of an industry operating at under 40% of its digital potential.

Average overall digital readiness score: 38.1 out of 100.

The gaps are structural:

• 77.1% of properties have no booking engine on their website
• 60.8% have no schema markup of any kind
• Only 7% implement Hotel or LodgingBusiness schema
• Only 1.5% have FAQPage schema
• 0% have dedicated room listing pages, dining pages, or amenities pages
• Median content completeness: ~41% — most properties tell less than half their story

Of 1,337 tested properties, only 76 (5.7%) were detected in any AI model response. The vast majority of accommodation websites are completely invisible to AI-powered hotel search.

What AI-Cited Hotel Properties Do Differently

The 76 properties that AI does cite share a consistent profile — and it challenges common assumptions about what drives AI visibility for hotels.

We measured the effect size (Cohen's d) of every factor that differentiates AI-cited properties from the rest — the higher the number, the bigger the measurable gap between properties AI cites and those it ignores:

1. URL quality (d = 0.60) — The largest effect. Clean, descriptive, human-readable URLs.
2. Trust signals (d = 0.50) — Cancellation policies, team pages, review integration, certifications.
3. Content quality (d = 0.36) — Clear, factual, entity-rich writing with specific and verifiable claims.
4. Technical SEO (d = 0.29)
5. Booking funnel (d = 0.26)
6. Schema completeness (d = 0.23)
7. Content completeness (d = 0.18)

The key finding: AI models favor properties that write clearly, build trust, and structure URLs well — not those with the fanciest websites.

Schema markup — which dominates the AI search optimization conversation — shows the smallest effect size in our data. This doesn't mean schema is unimportant; implementing Hotel/LodgingBusiness schema puts you in the top 7% of all properties, which is a valuable structural advantage. But if you're choosing where to invest your next hour, rewriting your property description with specific, factual, entity-rich language has a bigger measurable impact than adding JSON-LD.

A well-written 10-page hotel website beats a flashy 50-page one.

Five Countries, Five AI Visibility Fingerprints

Each of the five markets produces a structurally different AI citation landscape:

Austria is the most balanced market — lowest OTA dependency (22.3%), highest direct booking score (12.8%), strongest DMO ecosystem (Austria.info and Wien.info achieve 10.3% citation rates). Austrian properties also score highest on trust signals.

Croatia has the highest OTA dependency (30.7%) and lowest direct booking score (9.2%). When AI recommends accommodation in Croatia, it sends travelers to Booking.com more than in any other market. Croatian properties need stronger direct content, particularly in English and German.

Germany has the largest sample (540 properties, 39.1 average score) and the best booking funnel adoption — but schema completeness sits at just 1.8%. Most German operators still see their website as a technical requirement, not a strategic asset.

Slovenia is the most concentrated market (top 10 domains hold 33.3% of citations) but its DMOs punch above their weight (10.4% citation rate). Three independent Slovenian properties crack the top 10 most-mentioned companies in the entire dataset.

Switzerland has the highest STR platform citation rate (11.4%), reflecting its strong vacation rental market.

The Direct Booking Opportunity in AI Search

The "direct booking score" measures what percentage of AI URL citations point directly to a hotel or property website rather than routing through an intermediary.

Properties with a booking engine score 25% higher on digital maturity than those without (45+ vs. 36.3). And independent hotels actually receive more URL citations (11.8%) than hotel chains (4.3%), suggesting that distinctive, well-contented properties can compete for direct AI traffic.

But the median direct booking score for Gemini and Perplexity is 0.0% — meaning in more than half of all responses, not a single URL points to a direct hotel website. GPT-5's median of 15.4% is healthier.

The visibility formula from our data: distinctive identity (unique concept, design, or heritage) + strong third-party authority (TripAdvisor, Wikipedia, editorial coverage, DMO listings) + intent-aligned positioning (luxury, wellness, or business rather than generic accommodation).

No independent hotel in this dataset achieved significant AI visibility through website optimization alone. External validation is the primary driver of AI brand recall.

AI Search Optimization for Hotels: What To Do Now

The report includes a prioritized 90-day action plan. Here are the highest-impact actions, derived directly from the dataset:

Today

• Complete your Booking.com listing (95.3% of queries) and TripAdvisor profile (#2 cited domain)
• Register with your national and regional DMO (free, high citation yield)
• Check that robots.txt allows AI crawlers: GPTBot, GoogleExtended, ClaudeBot, PerplexityBot

This Week

• Build or sharpen your unique positioning in factual, specific language — vague USPs generate zero AI recall
• Create intent-specific landing pages targeting luxury, wellness, or business travel
• Ensure your website is available in your highest-value source languages

This Month

• Implement Hotel/LodgingBusiness schema (1–2 hours, joins the top 7%)
• Build event-specific landing pages if you're near a major venue
• Add dedicated room listing, dining, and amenities pages (0% of audited sites had these)

Starting Now

• Test 10–20 prompts a potential guest might ask AI about your location and property type
• Record whether you're mentioned, cited, or absent
• This is your baseline — repeat quarterly

Expert Perspectives on Hotel AI Visibility

The report includes contributions from four industry specialists:

Niclas Aunin (ALLMO.ai) provides a measurement framework for AI visibility, built on four pillars: technical clarity, content relevance, citable writing, and external authority. His mentions-versus-citations diagnostic matrix helps properties identify whether they're invisible, OTA-dependent, or optimally visible.

Sophia Milewski (Smartness / SmartPricing) argues that dynamic pricing is no longer just a revenue strategy — it's an AI trust signal. A fluctuating price is "proof of life" to algorithms. A static price list is perceived as potentially obsolete, leading to de-prioritization.

Domagoj Davidović (MinusMinus Agency) bridges the tension between branding and AI optimization, mapping AI's structural dimensions (Persona, Task, Context, Format) to branding fundamentals (Identity, Positioning, Channel, Tone of Voice). Creativity lives where humans decide; utility lives where machines rank.

Arne Petersen (VRM Days) diagnoses the German market's structural dependency on OTAs, arguing that success removed the urgency to change — and that most operators see AI as a productivity tool, not a distribution channel.

Prepared by Nokumo Research | NOKUMO SERVICES d.o.o. (Infranet Group)

This research is published under Creative Commons Attribution 4.0 International (CC BY 4.0). When sharing, cite Nokumo as the source and include the Nokumo logo.

With contributions from: Dario Alfirević (Chief Researcher), Tihana Alfirević (Editor), Niclas Aunin (ALLMO.ai), Sophia Milewski (Smartness), Domagoj Davidović (MinusMinus Agency), Arne Petersen (VRM Days).