Why Booking.com Gets Cited by AI and Your Website Doesn't

A global audit of 121,425 hotel homepages found that 36.3% carry no structured data at all, and of the ones that tried, 41.1% used the wrong schema type. Fewer than a third of hotel websites on earth are describing themselves to a machine correctly. Booking.com, meanwhile, has been doing this at scale for every property it lists for years. That gap, not any secret ranking factor, is most of the reason an AI model reaches for Booking.com's page instead of yours.

Getting cited by an AI model isn't about writing better marketing copy. It's about being legible to something that doesn't read the way a person does.

What Makes a Page Citable, and What Booking.com Has That Yours Probably Doesn't

An AI model synthesising a hotel recommendation isn't browsing your website the way a guest would. It's extracting structured facts: name, category, amenities, room types, ratings, coordinates, policies. Booking.com's per-property pages carry all of that in a dense, standardised, multilingual format, updated constantly, loading fast. That's precisely the profile a model treats as a reliable source. Most hotel websites don't offer anything close.

The Schema Gap, By the Numbers

Nicolas Sitter's March 2026 crawl of hotel homepages across seven countries put a number on the problem. Among 121,425 sites checked, 36.3% had no structured data whatsoever. Of the 55.8% that did implement JSON-LD, 41.1% used the wrong schema type entirely, tagging a hotel page as a generic LocalBusiness or leaving out the accommodation-specific fields that actually help a model understand what it's looking at. Run the maths and fewer than a third of hotel websites globally have correctly typed schema. A separate European accommodation audit of 1,337 sites found an even starker figure: only 7% carried proper Hotel or LodgingBusiness markup.

Which Schema Types Actually Move the Needle

Not every schema type carries the same weight, and not every AI engine parses them the same way. The table below maps what each type enables and which engines lean on it most.

What Each AI Engine Actually Wants From Your Content

Schema and entity consistency are the foundation, but each engine weighs signals a little differently once that foundation is in place. Building content that satisfies all five at once is possible, and it starts with knowing what each one is actually looking for.

Google AI Mode is worth a second look on its own. It cites an average of 25 sources per hotel travel answer, against 13.7 for ChatGPT, 8.0 for Gemini and a surprisingly low 0.8 for Perplexity, which appears to lean on model knowledge over live retrieval for this category more than the others do. For a hotel, AI Mode is the surface where having your facts right in as many places as possible matters most, simply because it is reading from more places at once.

Photography Is Structured Data Too

Multimodal search adds a layer most hotel marketers haven't thought about yet. Google AI Mode integrated Lens-based visual search in 2025 and expanded multi-object recognition within a single scene in March 2026. A traveller can photograph a pool they saw on Instagram and ask "where is this, and how do I book it?" Google Lens processes more than 20 billion visual searches a month globally, and hospitality is squarely inside that volume.

For a property, this means photography needs the same discipline as schema: images geotagged with accurate coordinates in their EXIF data, published with descriptive alt text that uses the property's official name and location, kept consistent across the hotel website and every OTA listing, and hosted at high resolution on the property's own domain where AI crawlers can actually index them. A stunning photo with no metadata connecting it to your entity sends a visual search to whichever competitor labelled their images correctly.

A llms.txt file is a plain, structured index of a website's key pages, written specifically for AI crawlers rather than for search engines. A March 2026 crawl of 105,002 hotel websites found only 6.3% had one. That figure matters less as an adoption statistic and more as a proxy: hotels with a llms.txt file scored 62% higher on general schema.org quality, meaning the file itself tends to travel with a property that has already taken its technical foundation seriously. Publishing one is a low-cost signal that also happens to correlate with everything else being in better shape.

Entity Consistency: The Silent AI Citation Killer

Schema tells an AI model what your property is. Entity consistency tells it that every source agreeing about your property is actually talking about the same place. The same official name, coordinates, phone number, category and policy text need to appear identically across Google Business Profile, Apple Maps, Wikidata and every OTA listing you hold. A mismatch, "Villa Kalyana Canggu" on one platform and "Kalyana Villas Canggu" on another, forces a model to guess whether it's looking at one property or two.

Cloudbeds found that nearly half of hotel brands in their 2025 dataset were misclassified by at least one AI platform. That's not a niche failure mode. It's close to a coin flip, and inconsistent entity data is the most common cause.

Fixing This: Where to Actually Start

1. Complete the OTA Listings First, Because They're Free

Before touching your own website's code, get Booking.com, Agoda, Traveloka and TripAdvisor to 100% completion. This costs nothing and puts you directly in the source layer every model already trusts.

2. Implement Correctly Typed Schema, Not Just Any Schema

Hotel or LodgingBusiness as the base type, with amenityFeature, Room, Offer, aggregateRating and FAQPage layered on top. Generic LocalBusiness markup, which is what most of the 41.1% mistyped sites default to, leaves out every accommodation-specific field a model can actually use.

3. Publish a llms.txt File

A short, curated index of your key pages, written for machines rather than for search engine crawlers in the traditional sense. It typically covers your homepage, room and rate pages, policies, and a handful of destination or amenity pages worth surfacing. It costs an afternoon of a developer's time and puts you in the 6.3% that has bothered. Given how strongly llms.txt presence correlates with overall schema quality in the March 2026 crawl, treat it as a forcing function: writing the index tends to expose exactly which pages are thin, outdated or missing the structured data this whole exercise is about.

4. Audit Entity Consistency Across Every Platform You're Listed On

Name, address, phone, category and policy text, checked line by line across Google Business Profile, Apple Maps, Wikidata and every OTA. Fix the smallest-looking discrepancies first, since they're usually the ones nobody noticed: a missing suite number, a phone number one digit off after a renumbering, a category set to "Guest house" on one platform and "Boutique hotel" on another. None of these look serious individually. Collectively, they're exactly the kind of noise that produces the misclassification Cloudbeds measured across nearly half of hotel brands.

5. Build a Review Programme That Rewards Specificity

Ask guests precise follow-up questions rather than open-ended ones. "What did you particularly enjoy about the pool area?" produces a citable sentence. "How was your stay?" produces "great, thanks," which is worth nothing to a model looking for something concrete to extract. Volume compounds this: a property with two hundred specific, positive reviews outweighs the one detailed complaint an AI model might otherwise latch onto.

Why This Compounds Instead of Resetting Each Month

None of these five fixes are one-time projects with a fixed end date. Schema needs revalidating every time room types or policies change. Entity data drifts as staff turn over and forget the naming convention. Review specificity depends on an ongoing prompt in your post-stay email sequence, not a single campaign. The properties that treat this as continuous maintenance, the same way they'd treat housekeeping standards, are the ones still showing up accurately in AI answers a year from now. The ones that fix it once and move on tend to drift back toward the 41.1% mistyped-schema group within a couple of platform migrations.

Frequently Asked Questions

What is schema markup and why does it matter for AI citation?

Schema markup is structured code on a webpage that describes what the page is about in a format machines can parse directly, rather than inferring it from prose. For hotels, correctly typed Hotel or LodgingBusiness schema with amenities, rooms and ratings gives an AI model the exact facts it needs to cite the property accurately.

What is llms.txt and do we really need one?

It's a plain-text index of a website's key pages, written specifically for AI crawlers. Only 6.3% of hotel websites globally have one, and those that do tend to score meaningfully higher on overall schema quality, making it a useful low-cost signal of technical maturity.

Our schema already uses JSON-LD. Are we covered?

Not necessarily. A 2026 audit found that of hotel sites using JSON-LD, 41.1% had the wrong schema type, most often a generic LocalBusiness tag instead of Hotel or LodgingBusiness. Having JSON-LD present is not the same as having the correct type with the accommodation-specific fields a model needs.

How do we fix entity inconsistency across platforms?

Line up your official name, address, phone number, category and policy text across Google Business Profile, Apple Maps, Wikidata and every OTA listing, and correct any discrepancy, however small. Nearly half of hotel brands in one 2025 dataset were misclassified by at least one AI platform, and inconsistent entity data is the most common cause.

Which AI engine should we prioritise optimising for?

There isn't one answer. Gemini and Google AI Overviews weight structured data and E-E-A-T signals heavily. Perplexity favours current, dated statistics in short extractable sections and has the highest review citation share of any model tested. Optimising for structured data, freshness and specificity tends to help across all of them at once.

This is exactly the kind of entity and structured-data work Tessar Napitupulu covers in depth in Cited or Silent, including a full platform-by-platform playbook for ChatGPT, Gemini, Perplexity, Copilot, Meta AI, Claude and Grok. Our Hotel GEO service runs this exact audit and fix sequence for individual properties, alongside the destination content work covered in our Hotel SEO service.