What 9,500 Hotel Website Scans Reveal About AI Visibility

AI-powered search tools — ChatGPT, Perplexity, Claude, Google AI Overviews — are now a meaningful part of how travellers discover and shortlist hotels. These tools do not read your website the way a human does. They crawl, parse structured data, read your robots.txt, and look for signals that communicate authority, completeness and trustworthiness.

The question we set out to answer was simple: how visible are hotel websites to these systems today? The answer, across more than 9,500 scans, is: not very. But the problems are well-defined, and the opportunities are real.

1. AI Bot Access: The Most Widespread Problem

Before a single piece of content, schema or metadata can be evaluated, there is a fundamental question: can AI crawlers even reach the site? The answer, for most hotels, is no — or only partially.

• 64% of hotel websites block at least one major AI bot in their robots.txt
• 18% block all four principal bots: GPTBot, ClaudeBot, Google-Extended and PerplexityBot
• Only 36% are fully open to all four

This matters more than any other single factor. In AIscore's methodology, bot access functions as a gate multiplier. A site that blocks three out of four crawlers and has excellent structured data will still score poorly — because the AI systems it is blocking will never see that data at all. Partial blockage is not a minor issue; it systematically suppresses the final score regardless of everything else.

The root cause is almost always benign: a web agency added a blanket Disallow: / for unknown user agents as a precaution, or a security plugin applied an overly broad ruleset years ago. Nobody intended to block AI bots — the category simply did not exist when the rule was written. But the effect is the same.

2. Schema.org: Present but Incomplete

Structured data is the single most powerful signal a hotel website can send to an AI system. JSON-LD Hotel schema tells an AI crawler exactly what the property is, where it is, what it costs, what facilities it offers and when guests can check in. Without it, the AI must infer — and inference is unreliable.

• 48% of hotel websites have some form of Hotel schema markup
• Only 22% include all four critical properties: starRating, priceRange, amenityFeature and checkInTime
• 73% are missing three or more of these properties

A Hotel schema block that exists but omits key properties is better than nothing — but not by much. An AI system asked "what amenities does this hotel have?" or "what is the check-in time?" cannot answer from structured data if those fields are missing. It must either infer from body text (error-prone) or decline to answer (worse).

The implementation gap is particularly striking given that platforms like WordPress, with plugins such as Yoast SEO or RankMath, make schema generation relatively straightforward. The bottleneck is not technical difficulty but awareness: most hoteliers and their agencies do not know that incomplete schema is a problem distinct from absent schema.

3. llms.txt: The Missed Opportunity

The llms.txt standard — a plain-text file placed at the root of a domain, structured to guide large language models towards the most important content on a site — was formalised in late 2024 and has already been adopted by a growing number of publishers, SaaS companies and e-commerce sites.

Among hotel websites, adoption is minimal: only 9% have an llms.txt file. The other 91% are leaving a meaningful differentiator on the table.

The value is straightforward. An AI model that encounters an llms.txt file receives a curated, authoritative summary of the site's key pages, contact information, booking links and property highlights — all in a format optimised for machine parsing. Without it, the model must piece together the same information from multiple pages, with higher risk of gaps or errors.

Creating a basic llms.txt file takes approximately one hour. For early adopters in a competitive hotel market, it is a measurable and durable advantage.

4. Meta Descriptions and Content Structure

The fundamentals of web visibility — clean content structure, descriptive metadata, well-formed headings — matter for AI systems as well as for traditional search. The data reveals widespread problems here too.

• 47% of hotel websites have suboptimal meta descriptions: either missing, too short (under 120 characters) or too long (over 160 characters)
• 23% have no H1 heading on the homepage
• 31% have multiple H1 headings — a structural error that confuses both search engines and AI crawlers about what the primary topic of the page is

H1 issues are particularly common on sites built with page builders like Elementor or Divi, where the visual design and the underlying semantic structure can easily diverge. A heading that looks like an H1 on screen may be a styled <div> or an H2 in the source code.

Meta descriptions are often neglected on inner pages — room category pages, restaurant pages, spa pages — even when the homepage is well-optimised. AI systems crawl the whole site, not just the homepage, and incomplete metadata on conversion-critical pages is a recurring pattern in the lower-scoring sites in our dataset.

5. Hreflang: An Underrated Problem

Hotels are international businesses. Most hotel websites serve audiences in multiple languages, and many have multilingual versions of their content. Hreflang attributes tell search engines and AI crawlers which version of a page corresponds to which language and region.

Among multilingual hotel websites in our dataset, 61% have hreflang errors — missing tags, misconfigured return tags, or inconsistent implementation across page types. This is a substantial proportion, and the downstream effect is real: an AI system that cannot reliably identify which version of a page is authoritative for a given language may default to the wrong version, produce mixed-language responses, or simply deprioritise the site.

Hreflang is one of the more technically demanding items on the optimisation checklist, but on WordPress with WPML or Polylang, the configuration is largely automated once set up correctly.

6. Grade Distribution Across 9,500+ Sites

Taken as a whole, the score distribution paints a clear picture of an industry that has not yet engaged with AI visibility as a measurable discipline:

• A (80-100): 4% of sites
• B (65-79): 11% of sites
• C (50-64): 20% of sites
• D (35-49): 28% of sites
• F (0-34): 37% of sites

65% of hotel websites score D or F. Fewer than one in six scores B or above. The median hotel website is essentially invisible to the AI systems that a growing share of its potential guests are using to make booking decisions.

The breadth of the improvement range reflects the diversity of starting positions. A site that is blocking all AI bots and has no schema can see dramatic gains quickly. A site that already has clean schema and open bot access will see more modest but still meaningful improvements from content structure and hreflang work.

The Opportunity Is Clear

The problems revealed by 9,500+ scans are not obscure or technically arcane. Robots.txt rules can be corrected in minutes. llms.txt files can be written in an afternoon. Schema completeness can be audited and improved within a week. H1 structure can be fixed during a routine content review.

The barrier is not complexity — it is awareness. Most hoteliers do not know their AI visibility score. Most do not know they are blocking the crawlers that power the tools their guests are using. Most have never heard of llms.txt. The information asymmetry between where the industry is and where it needs to be is the real problem, and it is one that is entirely solvable.

The hotels that close that gap first will not just score better on a benchmark. They will appear more reliably, more accurately and more completely in the AI-generated responses that are increasingly shaping where travellers choose to book.