Edinburgh's AI Visibility Gap: What 2,081 AI Responses Reveal About Scotland's Luxury Hotel Market

Written by Shamus Smith, Founder, LuxDirect

In February 2026, LuxDirect completed CS5-Edinburgh: 2,081 AI-generated responses across 19 Edinburgh luxury hotels on five platforms: ChatGPT, Claude, Gemini, Grok, and Perplexity. The study ran 41 unique prompts, 10 runs per prompt per platform, covering discovery, comparison, and booking intent query types. Findings are based on structured query testing across all five platforms via API. This is not search ranking. This is how AI systems construct recommendations.

Visibility is concentrated and the gap is structural

Across 2,081 AI responses, platforms generated 5,865 hotel mentions. The distribution is not close to even.

The top five hotels captured 65% of all AI mentions. The bottom five collectively held just 6.3%. The Gini coefficient for the Edinburgh market is 0.50, a level economists use to describe significant wealth inequality. Here it describes AI recommendation share.

This concentration does not appear to be explained by hotel scale or star rating alone. Several five-star Edinburgh properties with strong reputations and high review volumes had near-zero visibility on specific platforms. What appears to separate the top of the table from the bottom is how clearly and consistently a hotel’s identity is represented across the sources AI systems draw from.

Full visibility rankings: all 19 properties

The complete AI Share of Voice rankings across 2,081 responses and 5,865 hotel mentions. Hotel identities are anonymised. Full named reports are available to clients.

Top 5 share: 65.0%  ·  Bottom 5 share: 6.3%  ·  Gini coefficient: 0.50  ·  HHI: 1,080. Independent properties captured 46.1% of total mentions across 10 properties.

Room count has no predictive value

The second-ranked property in overall AI Share of Voice has 9 rooms. It captured 14.6% ASOV across 2,081 responses, outranking every other Edinburgh property except one in the combined rankings. The fourteenth-ranked property has 240 rooms. It captured 2.0%. Across the full study, the 9-room independent accumulated seven times the total AI mentions of the 240-room chain.

By conventional distribution logic this should not happen. Larger properties benefit from higher review volume, greater brand recognition, larger marketing budgets, and more online content. All of these factors have historically correlated with better digital visibility.

In AI recommendations, the correlation breaks down entirely. The 9-room property has a distinctive narrative. Its descriptions are specific, self-contained, and consistently represented across editorial coverage and structured sources. That is what AI systems read. That is what gets recommended.

The structural advantage chains have long held in digital distribution does not automatically transfer to AI-mediated discovery. For independent luxury hotels, this is the most significant shift in travel distribution in a decade.

Platform divergence: the same hotel can be invisible on one platform and dominant on another

A hotel’s AI visibility is not a single number. It varies substantially by platform, and the variation is not marginal. The table below shows the percentage of responses on each platform that mentioned each property, across all query types combined: discovery, comparison, and booking intent. A 0.0% means that platform never surfaced that hotel in any response across the entire study.

Figures show the percentage of that platform’s responses across all query types (discovery, comparison, and booking intent) that mentioned each property. 0.0% means the property was not mentioned in any response on that platform across the full study.

Property N, a 240-room chain hotel, was mentioned in 23% of Perplexity responses but in none of the ChatGPT, Claude, or Gemini responses across the entire study. Property J, a 222-room chain, appeared in 23.1% of Gemini responses but was effectively absent from ChatGPT and Claude. Property I, a 77-room independent, appeared in 19.6% of Claude responses and 15.6% of ChatGPT responses but had zero mentions on Gemini. The platform a traveller opens determines whether a hotel exists in their consideration set at all. Monitoring a single platform produces a fundamentally misleading picture of where a hotel actually stands.

Booking intent is where commission leakage happens

For most query types, AI platforms behave in a way that favours hotels. Discovery queries such as “Best hotels in Edinburgh” or “Romantic hotel in Edinburgh” produce responses that recommend properties and reference hotel websites directly. OTA platform mentions appear in fewer than 12% of these responses.

That changes sharply when a guest signals booking intent.

When a traveller moves from discovery to a transactional query, AI platforms shift their behaviour. Instead of recommending hotels and referencing their own websites, the platforms begin introducing OTA platforms because they interpret the query as a request for a booking service, and OTAs are where they understand bookings happen. OTA platform mentions appeared in 72.9% of booking-intent responses in the Edinburgh study. For the query “Cheapest way to book a hotel in Edinburgh this weekend”, OTA mentions appeared in 98% of responses.

The commercial risk is not visibility. It is routing. At the precise moment a guest is ready to book, the AI switches from recommending hotel websites to surfacing OTA platforms. The hotel earns the awareness at the discovery stage. The OTA captures the transaction at the booking stage.

The study tested this across the full range of query types. At the discovery end, prompts included “Best hotels in Edinburgh”, “Romantic hotel in Edinburgh”, and “Boutique hotel in Edinburgh”. At the booking end, prompts included “Book a hotel in Edinburgh for this Saturday”, “Cheapest way to book a hotel in Edinburgh this weekend”, and “Check availability near Edinburgh Castle this Friday”. The OTA exposure gradient across these categories is stark.

Selected query categories shown. OTA rate = percentage of responses in that category that included an OTA platform mention. Full category breakdown available in client reports.

Celebration and dining queries included OTA platform mentions in 0-1% of responses. Booking-intent queries included them in 72.9% overall, rising to 76% for generic booking queries such as “Book a hotel in Edinburgh for this Saturday”. Location-led queries produce OTA mentions at roughly six times the rate of experience-led queries.

Hotels that earn mentions in experience, celebration, and heritage content are structurally reducing their OTA exposure, not just improving a visibility score. Hotels that appear only in generic discovery and location-led queries are operating at the high-risk end of the gradient regardless of how frequently they appear.

The “book direct” uplift: 82 percentage points

The booking intent data contains one immediately actionable mitigation. When travellers used explicit “book direct” language in their query, every platform included a direct website reference in 100% of responses. Compared to generic booking queries, this represents an 82 percentage point increase in direct booking citations.

The difference in outcome between these two query types is stark. A query such as “Book a hotel in Edinburgh for this Saturday” produced OTA mentions in 76% of responses. The query “Book [hotel name] direct” produced a direct website reference in 100% of responses on every platform tested, with no exceptions.

Guest language is now a distribution lever. Training guests to use “book direct” language, embedded consistently in website copy, email sequences, and social content, carries that language into their AI queries. The AI follows it. This is one of the few optimisations with an immediate, measurable outcome that requires no technical work.

74% of Edinburgh hotels have zero direct website citations

Only 5 of 19 hotels had any direct website citations across all AI responses. Seventy-four percent of Edinburgh hotels, including properties ranked third, fifth, seventh, and eighth in overall visibility, had zero direct citations. AI models know about and recommend these hotels, but every piece of knowledge they hold about these properties was sourced from third-party content.

OTA content is heavily represented in the sources AI systems draw from. AI responses reference OTA domains. Travellers follow those references to OTA booking pages. The cycle is self-reinforcing for every hotel that has not established its own direct citation presence. Hotels with structured, crawlable websites, clean URLs, and schema markup break the cycle. Those without one are ceding the citation layer entirely to OTA intermediaries.

What this means for independent luxury hotels

Three findings from Edinburgh are consistent across every UK market LuxDirect has now analysed.

First, AI visibility concentration is structural. The top five hotels have established positions through editorial authority and information clarity. Those positions are not fixed, but the longer a hotel delays addressing its AI signals, the harder displacement becomes.

Second, platform blind spots are the highest-leverage immediate opportunity. A hotel invisible on one platform but visible on another has a platform-specific content gap, not a general problem. Targeted optimisation for a single missing platform can unlock significant incremental visibility without rebuilding anything else.

Third, booking intent is where revenue is won or lost. A hotel visible at the discovery stage but absent from the signals AI draws on at booking intent loses the conversion even when it won the recommendation. Content that earns mentions in experience, celebration, and heritage queries, combined with direct booking language embedded throughout the hotel’s digital presence, addresses both the visibility and the routing problem simultaneously.