Connected Isn't Chosen: The Three Layers of AI Travel Distribution

You can be connected to every AI platform in travel and still be invisible. That sentence would have sounded like a contradiction two years ago. Today it is the most important thing a distribution team can understand, and almost nobody is measuring it.

To see why, it helps to go back to a problem the industry already solved once, without realizing what it was actually solving.

The layer everyone already forgot

For years, the standard way OTAs and B2B distributors compared the same hotel across suppliers was by the price of its cheapest room type. Clean, simple, apparently apples to apples.

Except it wasn’t. A supplier could hold the exact room a traveler needed, at a genuinely good price, and never get shown, because the comparison ran on the cheapest room type rather than the right one. The offer was fine. The logic that decided what to surface was the whole game.

The suppliers who lost blamed their rates. Then their pipeline. Then their room mapping. They never saw the comparison logic itself, because nobody was measuring it. The sharp ones, usually the ones who listened to their engineers, figured it out and worked it: drop the margin on the cheapest room type to win placement, then make it back on the rest.

Call that Layer 1. Content, and the logic that ranks it. It has existed for as long as travel has had inventory, and it already taught the industry a hard lesson: the quality of your offer means nothing if the mechanism deciding what to show never surfaces it.

The layer everyone is celebrating

This week the industry is excited about the layer above it.

Hilton opened a direct line to Navan, giving a TMC direct API access to its reservation system with no GDS in the middle. Navan shipped a Model Context Protocol server so any AI tool can query its travel and expense data in natural language. These are real advances. They determine whether an AI system can reach you at all.

Call that Layer 2. Connectivity. Protocols, APIs, integrations, the plumbing that makes a brand technically accessible to an AI system. It is new, it is genuinely hard, and right now it is where most of the investment and most of the celebration is going.

The problem is that being reachable and being chosen are not the same thing.

The layer nobody is working on

The same week those connectivity wins landed, a Skift reporter ran a quiet experiment. She tested ChatGPT with Booking.com, Expedia, and Viator connectors already integrated. The chatbot told her they weren’t available and returned generic web results instead. Only after she pushed back did it surface the very connectors that were sitting there the whole time.

The integration wasn’t broken. The model’s behavior was.

Call that Layer 3. Model behavior. Even when you are fully connected, the model still decides whether to surface you, and that decision is driven by how the model was trained and how the prompt framed the request, not by whether your API responded. This is the layer that now determines who actually gets booked, and it is the one almost no travel company has a team for.

Why solving one layer doesn’t save you

Here is the part that should keep distribution leaders up at night. Each layer you solve reveals a harder one above it.

You can have perfect content. You can win every connectivity deal. And you can still lose, because the model chose not to show you. The suppliers who understood the ranking mechanism early won ten years of favorable sort order. The ones who blamed their rates paid for that decade without ever knowing why.

The asymmetry is in the skill set. Travel companies built teams for Layer 1. They are hiring for Layer 2 right now. Layer 3 requires understanding how models are trained and how they reason, which is a different discipline than API integration, and one most travel organizations simply do not have in the building yet.

Where this actually leaves you

This is not a story about AI being unreliable. It is a map of where the work is.

If your roadmap this quarter is entirely connectivity, you are solving a real problem and a solved-enough one. The open question, the one that will decide distribution for the next decade, is Layer 3: when an AI agent reasons about a trip, does it have a reason to surface you that it can actually read and act on?

That layer is workable. There are approaches that genuinely change how a model treats you. But they require people who live in both worlds, travel distribution and how these models actually behave, and who know how to turn one into leverage on the other.

This is exactly what we spend our days on while building Bitravel: an AI agent for corporate travel, built by people fluent in both the operational reality of distribution and how these models reason. Book a 30-minute call and we’ll walk you through what winning Layer 3 actually looks like, live, no deck.