OTAs are moving to own AI hotel discovery. Your property is funding the data that feeds it.

TLDR

• Lighthouse research presented at Luminate 2026 found that 82% of AI-generated hotel recommendations draw on OTA listings and editorial content, versus 18% from hotel-direct websites
• Booking Holdings and Expedia Group executives are reported to be positioning their companies to own the AI conversion layer, the step where a chat answer becomes a completed booking
• Three things to do this quarter: audit where your property shows up in AI search, add structured data for AI systems to read, and own your description layer on every channel

Lighthouse research presented by Blake Reiter, Lighthouse's Director of Hospitality Research, at Luminate 2026, found that 82% of AI-powered hotel recommendations pull from OTA listings and editorial review sites, not hotel-direct websites. The finding landed the same week trade coverage reported that Booking Holdings and Expedia Group executives are positioning AI discovery as an OTA play. That means the platforms want to own the conversion layer of how guests find and book hotels through ChatGPT, Perplexity, Google Gemini, and similar tools.

Booking.com and Expedia are positioning to own AI discovery. Your property data feeds it, but they control the link.

The OTA pitch to investors is straightforward. AI discovery tools will eventually replace a meaningful share of traditional hotel search. OTAs already hold the structured inventory data, user reviews, and conversion infrastructure AI systems need. A prompt like "boutique hotel in Edinburgh with parking under £150" becomes a completed booking through that infrastructure. Hotels supply the property descriptions, images, availability, and rates. The OTA keeps the relationship with the AI platform, and the commission when the guest books.

The precedent is Google Hotel Search. Hotels fund their Google Business Profiles and keep them updated. Google harvests that structured data and directs the booking through its own metasearch layer, or through an OTA partnership. Independent hotels built the content layer and lost control of the conversion layer. The AI shift is the same pattern, just moving faster.

82% already means the fight for AI discovery is more than half lost

Reiter's 82% figure is the tell. AI models trained on public web data favor OTA pages. Those pages are built for search engines, carry more inbound links, and aggregate reviews at scale. A hotel's own brand.com page, however well designed, sits lower in the training data. It also ranks lower in retrieval, whenever an AI system answers a hotel question. Independent hotels are already fighting uphill in how AI systems learn to describe them.

Airbnb hosts learned a version of this lesson in traditional search. Hosts built the listings, uploaded the photos, and kept availability calendars current, but Airbnb controlled the discovery layer and the guest relationship. Hotels are now in a similar position with OTAs in the AI layer, with less leverage than Airbnb's hosts had. At least Airbnb's hosts were platform-exclusive. Independent hotels list everywhere.

Three open questions the OTA framing doesn't answer

The Lighthouse research doesn't break out whether the 82% figure shifts by property type, destination, or AI platform. ChatGPT's hotel picks may lean more on TripAdvisor editorial than Perplexity's do. Boutique queries may pull more direct-site data than budget queries do. We'll revisit this once a platform-level breakdown exists.

Nor have OTA executives explained what happens if an AI system hallucinates an offer, invents an amenity, or quotes an outdated rate. Traditional OTA bookings flow through a structured API with defined liability. AI-recommended bookings will likely flow through the same pipes, meaning the OTA's terms of service still apply. No one has published error-rate data on AI-recommended bookings yet.

Perplexity and other AI search engines are smaller, but follow the same direct-link model

ChatGPT is the largest AI search interface by user base, but Perplexity, Google's Gemini, and Microsoft's Copilot follow the same outbound-link model. They cite sources and link directly to websites, rather than aggregating results into a middle layer. For independent hotels, that's structurally different from how Google search has worked for the last decade. Google sent traffic to aggregators. AI search sends traffic to the original source.

The catch is that AI search engines favor authoritative, well-structured content. A booking.com listing is well-structured by design. A hotel's own website might not be. AI doesn't reward old SEO tricks. It rewards a site that clearly answers the question the guest asked.

The framework that decides AI visibility: schema markup, description ownership, and presence testing

Three elements determine whether an AI system recommends a hotel directly or defers to an OTA. Schema markup is structured code on a website, using the Schema.org Hotel vocabulary. It tells search and AI systems a property's type, amenities, address, and rates, in a format machines can parse. Description ownership means the text describing a property on every channel. AI systems tend to quote whichever version of that text they find most specific and well-structured. Presence testing means checking, on a regular basis, whether a property shows up at all when someone asks an AI system for a hotel like it. OTAs have invested in all three for years. Most independent hotel websites have invested in none of them.

Branded chains have loyalty data. Independent hotels have neither loyalty data nor OTA scale.

Branded chains hold one advantage independents don't: a loyalty program's first-party guest data. Large chains are starting to feed that data into their own AI tools, rather than handing it to OTAs. Independent hotels have no equivalent asset. They can't out-spend Booking.com or Expedia on structured-data infrastructure, and they don't have a loyalty database to fall back on either. Their only lever is the one they've always had: a property description specific enough that an AI system has no reason to prefer the generic OTA version instead.

Three things to do this quarter: audit where you show up, build structured data, own your description layer

Test where your property appears in AI discovery today: ask ChatGPT, Perplexity, and Google Gemini for hotels in your city with your signature amenity, and check whether the answer links to your site or an OTA page.

Add Schema.org Hotel markup to your website: this structured-data format tells AI systems and search engines your property type, amenities, address, rates, and availability, the same way OTA listings already do.

Rewrite your property description on every channel: the text on Booking.com, Expedia, Google, and TripAdvisor is what an AI system will quote when it recommends your hotel, so make it specific rather than generic.