Google and Microsoft agreed on how an AI agent finds your hotel
Their new standard lets your hotel publish a list of what an AI agent may do with your rooms. It doesn't say whether the agent should use your list or your OTA's
by Markus Busch
The file. On June 17, Google and Microsoft announced a specification called Agentic Resource Discovery, alongside nine other companies. It asks you to put one small file at a fixed address on your own website — /.well-known/ard.json — listing what your property can do for a machine that shows up asking: check availability, quote a rate, take a booking, cancel one. Anyone may implement it, and it is still a proposal. The published text is version 0.91.
You have been asked for a file before. In July this publication took apart llms.txt, the file hotels were told to publish so the AI would understand them. Google's own documentation said Search doesn't use it, and server logs showed the machines never even asked for it. This is not that. llms.txt was a convention nobody adopted. This one is co-signed by the two companies that run the search indexes, with working code and a registry service behind it. The reason to look at it is not that it is new. It is who wrote it.
What the file changes. Your website tells a reader what your hotel is. The listing tells a machine what it may do to your hotel. That is a different kind of document. Your rate stops being a number on a page and becomes something an agent can act on. Your cancellation window stops being a paragraph in the terms and becomes a rule the machine applies. The hotel is no longer being described. It is being operated.
What the standard settles. Who you are, and it settles it hard. A listing that claims to speak for your website is thrown out unless whoever published it can prove, cryptographically, that your website stands behind it. The spec calls this its defense against squatting. Nobody can put up a file claiming to be your hotel and be believed.
What it refuses to settle. Which of the real ones wins. The standard has no rule for whose listing takes priority when several describe the same hotel, and it says the ranking a discovery service returns "MUST NOT be interpreted as a trust, compliance, or safety judgment." Microsoft is equally plain that the standard "doesn't replace authentication, authorization, governance, or organizational trust decisions." So picture your hotel, your booking engine, three OTAs and a wholesaler each publishing a verified listing for your forty rooms. Every one of them is genuine. Nothing in the standard says which one an agent should act on, and that gap is deliberate.
Somebody will fill it. In a piece this week arguing that hotels need the right technology partners, Mirai's chief executive for the Americas offered a candidate: authority, he wrote, may depend on reputation, data quality, freshness "and, potentially, the commercial or contractual relationship between the hotel and the technology provider." That is a distribution vendor's view, not a provision of the spec. It is also the most honest sentence written about this so far.
Monday. Ask whoever runs your website and your booking engine one question: do you intend to publish one of these for my property, under whose domain, and what will it say I permit? Eleven companies wrote this standard. Cisco, GitHub, Nvidia, Salesforce, Snowflake. Not one of them sells a hotel room.
The file takes an afternoon to publish. Whose name is on it is the whole of your distribution.
Read also: Your hotel is paying for an AI file that nothing reads
Enjoying this analysis?* Hospitality.today delivers daily insights on hotel distribution, AI trends, and travel commerce — straight to your inbox. Subscribe for free at Hospitality.today →