Vibe Booking: Hotel Data Is Not AI Ready. Here's Why

From the Travel Tech Podcast newsletter for episode 18, Vibe Booking: Hotel Data Is Not AI Ready. Here's Why.

I've been wanting to dig into the data plumbing underneath travel distribution for a while. One of our previous episodes, about a fake hotel site that successfully captured real bookings and left guests stranded at check-in, opened a thread I couldn't close: how does bad data cause that kind of breakdown at scale? For those of you across aviation, travel tech, and AI, I knew this needed to go deeper. This week I did.

My guest built two companies out of frustration with a problem he spotted the first day he walked into a large Miami travel agency: 20 agents with 10 supplier portals open simultaneously, comparing room rates by eye because the industry had no reliable way to match identical rooms across systems. He went from building antivirus heuristics to writing NLP-based room-matching engines, and eventually to solving the upstream data quality problem that makes all of it necessary. You can listen now on Apple Podcasts, also streaming on Spotify and YouTube.

What came out of our conversation wasn't just a technical explainer. It was a picture of an industry where the incentives actively resist the fix.

Travel distribution runs on a web of B2B API partnerships so interlocking that a hotel's own inventory can complete a full circuit: leaving one distributor, passing through several partners, and arriving back at the original supplier, marked up. This isn't a bug anyone is rushing to fix. The cash economics of the current model are good for the platforms and SaaS layers that sit in the middle. The cost is paid in data quality, confirmation errors, and the hours agents spend manually reconciling availability across portals.

AI is now turning this latent problem into an urgent one. LLMs don't query availability APIs the way a GDS terminal does. They scrape the web, ground themselves on what they find, and surface properties whose content is structured, specific, and trustworthy. A hotel with normalized, experience-rich data on its own website ranks higher in the emerging AI distribution layer. One that has pushed its inventory through five intermediaries and lost its personality along the way is effectively invisible to an AI agent trying to match a traveler's question to a property.

My guest is Olivier Boinet, founder of room-matching.com and Omnitravel.ai. We covered room deduplication at scale, the mechanics of data normalization loss, why the check-recheck-check confirmation loop is structurally baked in, how LLMs evaluate hotel websites as a first-pass filter, corpus-based personalization for different traveler profiles, and what Olivier calls "vibe booking": the idea that content quality is now a direct booking strategy.

Key Takeaways

• Room mapping is still largely manual in most agencies: Without dynamic matching, agents open 10 or more supplier portals simultaneously to compare rates on what are effectively the same rooms described in incompatible formats.
• Normalization destroys differentiation: Every intermediary layer a hotel pushes its data through strips out the specific offers, experiences, and language that distinguish it. What arrives at the end of the chain is a commodity record.
• Distribution partnerships create circular inventory: B2B daisy-chains are dense enough that a hotel will often re-encounter its own rooms through a partner that sourced them from another partner that sourced them from the hotel itself, each layer adding markup.
• LLMs treat hotel websites as grounding sources first: Properties that provide structured, high-quality, machine-legible content get weighted higher in AI recommendation outputs. The AI discoverability layer rewards content investment before a single query is even made.
• Corpus-based retrieval makes personalization fast and cheap: By organizing property content into audience-specific corpora, an AI agent can serve a dramatically different response to a Gen Z solo traveler versus a business booker without generating new content on each query. It just retrieves the right branch.

Alex Brooker, Host of the Travel Tech Podcast, Founder of Airside Labs