The AI recommendation is the new battleground for hotel distribution

At Luminate 2026, Lighthouse Director of Hospitality Research Blake Reiter took the stage with a simple but stark premise: AI has become a primary channel for hotel discovery, it operates with clear and measurable biases, and most hotels have no idea where they stand.

What followed was one of the most data-dense presentations of the day: a study built on over 4,500 ChatGPT prompts across nine global destinations and five traveler personas, designed to answer three questions that every hotel marketer should be asking right now.

What hotels is AI recommending? Why those and not others? And where does it all come from?

Here's what the data showed.

The shift that feels familiar

Most hotel marketers have lived through a version of this moment before. Not long ago, cracking Google was the distribution challenge. Hotels that figured out SEO early built compounding advantages that lasted years. What's happening now feels similar, except the timeline is compressed.

AI search is now doing what Google once did, acting as the first point of contact between a traveler and their next hotel.

Travelers are increasingly starting their travel planning with a question rather than a search. It used to be search, click, browse, book. Increasingly, it's ask, get a recommendation, book. That middle step, the AI recommendation, is the new battleground.

ChatGPT now has over 900 million users. Booking.com and Expedia are already embedded in its app store. When a traveler opens ChatGPT or another LLM (large language model) and asks where to stay in Barcelona, the answer shapes a real booking decision. If your hotel isn't in that answer, you're losing opportunities at a scale that's very hard to measure, because you're invisible.

What the research found

To understand how this actually works, Lighthouse ran a study using nine global destinations (across the US, Europe and APAC), five traveler personas (luxury, business, family, budget and generic) and 101 prompt variations per destination-persona combination. That produced 4,545 distinct prompts fed into ChatGPT, each using a clean independent crawler to eliminate bias.

The goal was simple: follow the data into the generative AI black box.

Most hotels are invisible by default

Across all 4,545 prompts, ChatGPT mentioned hotels by name 49,707 times. The number of unique properties in that pool? 2,721.

Everything else was a repeat. A small group of hotels is being recommended over and over, while the vast majority never appear at all.

Two metrics help frame the scale of this problem.

Market coverage rate measures what percentage of hotels in a given market ChatGPT actually surfaces. In Tokyo, that number is 10%. In Paris, 13%. Even in Park City, a small, highly seasonal market, two thirds of hotels were never mentioned across thousands of prompts.

AI invisibility is the default condition for most hotels. You have to earn your way out of it.

Share of voice measures the concentration of mentions within a market. In Paris, the single most-recommended hotel captured 3% of all mentions. Paris has roughly 2,000 hotels. One property, out of two thousand, was claiming that share.

The top 100 most-mentioned hotels globally accounted for over 13% of all mentions. The math is working against most properties.

Chains vs. Independents

The chain versus independent split is not flattering for independents. In Indianapolis, just 6.5% of hotel mentions went to independent properties. In Tokyo, 10.5%. Paris was the only market where independents exceeded chains in share of mentions, and even there, the gap was narrow.

The imbalance isn't just striking on its own. It becomes more significant when you compare it to actual supply. Brussels has significantly more independent hotels than chains, yet independent mentions came in at just 28%. Miami and Tokyo tell the same story.

Among branded chains, the US market is essentially Marriott's world. More than a quarter of all branded hotel mentions in the US went to Marriott, meaning if ChatGPT recommends a branded hotel, one in four times it's Marriott. Hilton came in a distant second, and the top three brand families combined accounted for 54% of all branded mentions.

Europe is more democratized: Accor and Marriott are effectively tied at around 11% each, and ten brands sit between two and five percent of mentions. Tokyo is the most equitable market studied: Hyatt leads at just 5.8%, Marriott doesn't appear until eighth place, and the top 20 brands together account for 64% of mentions (compared to three brands accounting for 54% in the US).

Los Angeles data illustrates the branded concentration most starkly. The Dorchester Collection, with just two hotels in the entire market, was capturing around 6% of all LA hotel mentions. Omni had one hotel in LA and was pulling 5%. Millennium had one hotel and got 3%. Brand visibility in AI vastly outpaces brand share of actual supply.

AI has a type, and it skews luxury

Guest review scores, it turns out, are not what drives AI recommendations. The correlation between a hotel's Booking.com score and its ChatGPT visibility was weak. Something else is driving selection.

Star rating tells a clearer story. Even on generic prompts, where travelers gave no guidance, no budget indication, nothing beyond "I need a hotel in X," four and five-star properties dominated. Three-star hotels were nearly absent from that discovery pool. And generic prompts represent the most persuadable travelers: people who haven't made up their minds yet.

The bias holds across personas. Business travel recommendations ran 83% four and five-star, despite three-star chains being entirely appropriate for most corporate stays. Family-friendly results came in at 73% four and five-star. The only persona where three-star properties got meaningful mentions was budget, and even there, the concentration was modest.

For mid-tier properties, this is a strategic wake-up call. Product tier is a headwind, but it's one that content language and digital positioning can actively work against. The machine learning that drives these models tends to reinforce existing patterns, which means properties absent from recommendations today are likely to stay absent unless something changes.How your hotel is described across every touchpoint shapes which travelers AI sends you, or whether it sends you anyone at all.

Your language shapes who finds you

Beyond star level, ChatGPT shows a clear preference for "play-led" properties: hotels in arts districts, near attractions, close to nightlife and shopping. In generic searches, more than half of all recommendations fell into this category.

The encouraging finding is that ChatGPT does respond to intent. Business prompts generated more recommendations for commercial-district hotels. Budget prompts surfaced more affordability-led properties. Luxury and family searches returned more nature-led results.

But here's the implication that matters: the language you use to describe your hotel across OTA listings, editorial coverage and your own website is actively teaching AI which travelers to send you. If a Paris hotel's listings emphasize business amenities and meeting facilities, it will appear in business-travel searches. If those same signals are absent, it won't, regardless of the actual demand in that market.

Your content is no longer just marketing copy. It's algorithm input.

Where AI gets its information and where it sends travelers

When AI assistants formulate a hotel recommendation, it cross-references sources to decide what to show. Of those sources, 82% come from two categories: OTA and metasearch platforms (Booking.com, Expedia and similar), and editorial and media sites (Forbes, Lonely Planet, Condé Nast Traveler and equivalent publications).

This has a direct implication for PR strategy. Getting your hotel mentioned in a "best of" list or editorial feature isn't just good for perception. It's now a distribution input. The line between PR and distribution strategy is dissolving. The same dynamic applies to Google's AI Overviews, where editorial mentions and structured data similarly determine which hotels surface.

What happens once ChatGPT makes a recommendation is equally important. When the research tracked clickable links, specifically the hyperlinks and map links that travelers actually click through, the picture shifted dramatically. OTA and editorial sites dropped off. The majority of outbound clicks went to hotel websites directly.

If AI recommends your property, it gives travelers an express lane to your own booking engine. The OTA doesn't take a commission on that journey. The guest lands directly on your site.

Which means the question isn't just whether you're discoverable. It's whether you're ready to convert the traffic when it arrives.

What to do now

Blake closed with five actions every hotel team should take. They're worth repeating:

• Treat PR and editorial coverage as a distribution channel. Being featured in credible publications and "best of" lists is now an AI data input, not just a brand signal

• Audit your OTA and metasearch listings. These are what AI uses to verify what it already knows. Stale, incomplete or inconsistent listings mean AI is working with bad data

• Audit your content language against your target guest. The words used across your website and digital touchpoints determine which traveler segment AI routes to you

• Know where you actually stand in AI today. Open ChatGPT, run searches with and without persona context, and see what comes back. Many hotels are trying to manage a distribution channel they've never looked at

Build for direct booking conversion. When AI does send travelers to your site, you need to be ready to close them

The measurement layer matters too

The research makes one thing clear: AI visibility isn't random. It's patterned, consistent and measurable, which means it can be influenced, optimized and tracked over time.

That's exactly what Lighthouse built Connect AI to do. It puts hotels inside AI travel conversations with verified brand content and live rates, and drives travelers directly to the hotel's booking engine rather than through an OTA. The GEO component ensures AI platforms understand your hotel the way you define it, not as third-party sources describe it.

AI Visibility Insights adds the measurement layer: the ability to track where you rank in AI recommendations, benchmark against your competitive set, and identify the content gaps that are costing you mentions. Think of it as the AI equivalent of tracking your Google rank, the same discipline you've applied to search for the past 20 years, now applied to the channel that's increasingly shaping where travelers go next, and ultimately the customer experience hotels can deliver.

Hotels that move first will shape what AI learns to recommend. The data tells you where the gaps are. Now it's about closing them.

The data goes even deeper. Watch Blake walk through the full AI Visibility Edge research from Luminate 2026: