How AI Engines Recommend Hotels: A 500-Query Study

The headline finding

Delphium Labs ran 500 real traveller queries across ChatGPT, Perplexity, and Gemini in January 2026. The single clearest result: chain hotels appear in 72% of AI-generated recommendations, while independent properties account for just 28%. That gap is not random. It is driven by specific, measurable factors in how properties present themselves online, and understanding those factors changes what hotels can do about their AI visibility.

This post breaks down the full methodology, the engine-by-engine differences, and the practical signals that determine which hotels AI engines choose to recommend.

How we ran the study

Delphium Labs conducted this analysis using a structured query set designed to mirror how real travellers ask AI engines for hotel recommendations. We built 500 distinct queries across five categories:

• City breaks (120 queries): "best hotel in Edinburgh", "where to stay in Bristol for a weekend", "affordable hotel near central London"
• Romantic getaways (100 queries): "romantic hotel Lake District", "boutique hotel for anniversary trip UK"
• Business travel (100 queries): "hotel near Birmingham NEC with good WiFi", "best business hotel Manchester city centre"
• Family holidays (80 queries): "family-friendly hotel Cornwall with pool", "kid-friendly accommodation near Alton Towers"
• Boutique and design stays (100 queries): "design hotel Liverpool", "unique boutique hotel Brighton seafront"

Each query was run across all three engines within the same 48-hour window (6-8 January 2026) to control for temporal variation. We recorded every property mentioned by name, its position in the response, whether a source was cited, and the type of property (chain, independent, or boutique group).

A property was classified as "chain" if it belonged to a group operating more than 20 properties. Everything else fell under "independent", including small boutique collections of two to five hotels.

The overall numbers

Across all 500 queries and three engines (1,500 total responses), Delphium Labs analysis found the following distribution:

• 72% of named hotel recommendations were chain properties
• 28% were independent properties
• The average response mentioned 4.3 specific hotels by name
• 61% of responses included at least one independent property
• Only 12% of responses featured a majority of independent properties

Those top-line figures tell one story. The underlying signals tell a more useful one.

What drives AI hotel recommendations

Delphium Labs identified six measurable factors that correlate with a property appearing in AI engine responses. We ranked these by the strength of their correlation across the full dataset.

1. Structured data and schema markup

Properties with comprehensive schema markup on their websites appeared at significantly higher rates than those without it. Delphium Labs analysis found that hotels implementing full HotelRoom, AggregateRating, and LocalBusiness schema were 2.1x more likely to be cited than comparable properties without structured data.

This was the single strongest technical signal in the study. Chain hotels benefit here because their corporate web teams implement schema at scale. Independent properties often lack it entirely.

2. Detailed room and facility descriptions

Properties with detailed, descriptive room content on their websites were cited 3.4x more often than those with minimal descriptions. This goes beyond listing "double room" or "suite". AI engines pulled from pages that described specific features: room dimensions, view descriptions, bathroom amenities, unique design elements.

A property page stating "24sqm room with floor-to-ceiling windows overlooking the harbour, rainfall shower, and Roberts radio" gives an AI engine concrete material to work with. A page listing "Standard Double - from 120 per night" does not.

3. Review volume over review score

Delphium Labs analysis found that review volume correlated more strongly with AI visibility than average review score. A hotel with 1,200 reviews and a 4.2 rating appeared more frequently than a hotel with 40 reviews and a 4.8 rating. The threshold effect was notable: properties with fewer than 100 Google reviews were significantly underrepresented across all three engines.

This makes intuitive sense. AI models are trained on and retrieve from large text corpora. More reviews mean more textual data associated with a property, which means more opportunities for that property to surface in model outputs.

4. Direct booking pages with transparent pricing

Hotels with clear, crawlable direct booking pages that displayed room rates appeared more often than properties whose pricing was locked behind OTA-only distribution. Delphium Labs found that properties with publicly visible rate information on their own websites were cited 1.8x more frequently than those relying entirely on third-party booking platforms for pricing.

This finding has a straightforward explanation. AI engines can crawl and index a hotel's own website. They cannot always access dynamic OTA pricing. If the only place a room rate exists in text form is behind an Expedia or Booking.com search interface, the AI engine has less pricing data to reference.

5. Google Business Profile completeness

A fully completed Google Business Profile, including photos, attributes, room types, and regular post updates, correlated with higher visibility across all three engines but most strongly with Gemini. Delphium Labs analysis found that properties with more than 50 Google Business Profile photos and completed attribute fields appeared 1.6x more often than those with minimal profiles.

6. Content depth and topical authority

Hotels that maintained blog content, local area guides, or detailed "things to do" pages on their websites showed higher visibility for contextual queries. When a traveller asks "romantic hotel near the Cotswolds with good restaurants nearby", the AI engine favours properties whose websites discuss the local restaurant scene. Content that positions a hotel within its local context performs better than content that describes the hotel in isolation.

Engine-by-engine breakdown

The three engines showed distinct patterns that matter for visibility strategy.

ChatGPT

ChatGPT showed the strongest bias toward well-known brands. In Delphium Labs testing, chain hotels accounted for 78% of ChatGPT recommendations, the highest of the three engines. ChatGPT rarely cited sources, making it difficult to trace why specific properties appeared. Its responses tended toward "safe" recommendations: properties with high name recognition, consistent quality reputations, and broad availability.

For independent hotels, ChatGPT visibility correlated most strongly with the property appearing in multiple authoritative travel publications. If a hotel had been featured in Conde Nast Traveller, The Guardian travel section, or similar outlets, it was significantly more likely to appear in ChatGPT responses.

Perplexity

Perplexity was the most favourable engine for independent properties. Independents appeared in 35% of Perplexity recommendations, compared to 28% across all engines. The key difference: Perplexity cites its sources explicitly, and it draws heavily from recent web content.

Delphium Labs analysis found that Perplexity favoured properties with strong, recently updated website content. A boutique hotel that had published a blog post about its new tasting menu in the past 30 days was more likely to appear than a chain property with a static website. For independents investing in content marketing, Perplexity is currently the highest-return engine.

Gemini

Gemini leaned most heavily on Google's own data ecosystem. Google Business Profile completeness was the single strongest predictor of Gemini visibility in Delphium Labs testing. Gemini also showed the strongest correlation with Google Maps data, review content from Google Reviews specifically, and Google Hotels pricing data.

For properties that invest in their Google Business Profile, Gemini offers the clearest direct path to AI visibility. The signal is more measurable and more controllable than the factors driving ChatGPT or Perplexity.

What this means for independent hotels

The 72/28 split is real, but it is not fixed. Delphium Labs identified specific actions that independent properties can take to shift their visibility:

Implement comprehensive schema markup. This is the highest-impact technical change. Full HotelRoom schema with room types, amenities, pricing, and images gives AI engines structured data to pull from. Most independent hotel websites lack this entirely.

Write detailed, specific room and property descriptions. Generic descriptions do not surface in AI recommendations. Specific, sensory descriptions of rooms, views, amenities, and unique features give AI engines material to cite. Every room type should have at least 150 words of unique descriptive content.

Build review volume deliberately. A post-stay email sequence that makes reviewing simple can shift a property from 80 reviews to 400 over 12 months. That volume change has a measurable impact on AI visibility.

Publish transparent pricing on your own website. If your rates only exist on OTAs, AI engines have less data to work with. A clear, crawlable pricing page or booking widget with visible rates improves your chances.

Invest in Google Business Profile as a primary channel. Complete every field, add photos regularly, post updates, and respond to reviews. This is especially important for Gemini visibility, but it affects all three engines.

Create local context content. Write about your area, your neighbourhood, the restaurants nearby, the experiences available. AI engines increasingly recommend hotels in context, and properties that provide that context on their own websites are better positioned.

What comes next

These findings shaped the development of FindingFin, the AI visibility tracking tool from Delphium Labs. The patterns we identified in this 500-query study, particularly around structured data, content depth, and engine-specific behaviours, informed how FindingFin monitors and scores property visibility across AI engines.

The data is clear: AI-driven hotel discovery is not a future trend. It is happening now, and the properties that understand its mechanics have a measurable advantage. The question for any hotel is not whether travellers are using AI to find places to stay. It is whether your property appears when they do.

Delphium Labs will continue publishing research from ongoing AI visibility studies. The next analysis examines the chain vs independent split in greater detail, looking at the specific query types where independent properties can and do compete effectively.