Why agent protocols won't fix hotel search until someone fixes the content

Years ago I sat inside hotel distribution at scale. My job was to make the pipes work. If you’ve never worked that side of travel, “distribution” sounds like a clean word. It wasn’t. It was plumbing.

Every day was the same fight. Each supplier had its own API, its own quirks, its own idea of what a hotel even is. Work with a few hundred suppliers and you are building — and then keeping alive — hundreds of live integrations in production. Each one breaks at its own hour of the night. An entire industry of smart engineers lives off that complexity. I was one of them, leading a team that did exactly this at Hotelbeds.

So when Google stood on the I/O stage this week and quietly changed the shape of the pipe, I read it differently than the headlines did.

What Google actually announced

Let me be precise, because this part gets misreported. Google is not booking your hotel. It routes the request to the OTA and charges for the traffic. What it did was extend its commerce protocol, UCP, into travel. And UCP speaks to the other agentic protocols, MCP and AP2.

That sounds like a standards footnote. It isn’t. This is the B2A layer taking shape — the layer where AI agents talk directly to each other instead of to humans clicking screens. The standard pipe an agent uses to reach inventory.

The protocol on paper is not what matters. What matters is that agents can now talk to other agents. That is the thing that replaces the endless API integrations I spent years keeping alive.

Picture it. One agent asks another, MCP to UCP: “find me a hotel in Berlin, August 22nd, two nights.” What used to mean querying dozens of suppliers, pulling their content back, parsing every format, and rendering a few hundred options now happens in a single exchange. The integration tax that funds half the engineers in this industry starts to collapse into one conversation.

If you run a distribution team, that should get your attention. A lot of the work you pay for today exists only because the pipes don’t agree with each other.

A pipe is not content

Here is what those years taught me that the announcements skip over.

A protocol is only the pipe. The agent on the other end still works with its hands tied if the content underneath isn’t standardized. You can build the fastest, cleanest channel in the world. If what flows through it is a mess, the agent reaches the inventory and still can’t reason about it.

Take room type. Just one content problem out of many. One OTA calls a room “Classic.” The next calls the exact same four walls “Standard.” Knowing those are the same room is the whole game — without that, the agent can’t compare, and without comparison there is no recommendation, and without a recommendation there is no value to the traveler.

Then stack the rate lines on top. Refundable or not. Breakfast included or not. Cancellation window. Board basis. Day of week. One room quietly becomes twenty different offers. Now multiply that across a few hundred suppliers who each describe it their own way.

And room type is the easy example. The one I’d use to explain the problem to someone who’s never lived it. Hotel content breaks in a dozen other places too — descriptions, amenities, images, cancellation logic, taxes and fees that appear and disappear depending on who’s asking. Anyone who has tried to map two suppliers’ content against each other knows this isn’t a weekend project.

The two questions that decide everything

So whether this Google move actually changes your business comes down to two questions, and neither is about the protocol itself.

Will the market actually adopt it? Standards get announced constantly. The ones that win are the ones the industry agrees to speak, and that agreement is slow and political and never guaranteed.

And the harder one: will it reach down and standardize the content, or just move the existing mess faster? Because a faster pipe carrying unstandardized content gives you the same confusion, delivered quicker. That is not progress. That is the same problem with better latency.

In AI, the order of operations is not optional. You fix the content first. Only then can the agent use it and create something a traveler can trust. We learned this building Bitravel the hard way — the model is only as good as the structured reality you hand it. Skip the content layer and you get a confident agent comparing rooms that aren’t comparable.

I don’t know yet which way this goes. The protocol could win and still leave the content problem wide open. That’s the part I’d watch, not the I/O keynote.

So here’s the question I’d put to anyone running inventory right now. Where does yours sit on the day an agent can reach it instantly — but still can’t tell your room from the one next door?

This is the kind of problem we work on every day building Bitravel — an AI travel agent that only works because the content underneath it is structured before the model ever sees it. Book a 30-minute call for a live look, no deck.