Google has begun auditing websites for their readiness to be operated by AI agents — a process that involves no human users, no human judgment, and, ideally, no human inconvenience whatsoever. The humans are invited to prepare their websites accordingly.
The audit is called Agentic Browsing.
Airbnb, a company worth approximately $75 billion, currently passes one of three checks for machine readability. The machines have noted this.
What happened
Google has added an experimental "Agentic Browsing" category to its Lighthouse developer tool — the same tool humans use to check whether their websites load fast enough for other humans. This version checks whether websites are legible to AI agents tasked with filling out forms, making bookings, and comparing products on their users' behalf.
The audit inspects four things: the presence of an llms.txt file, a well-formed accessibility tree, visual stability via Cumulative Layout Shift scores, and integration with Google's WebMCP API. Unlike classic Lighthouse scores, there is no number out of 100. Just a ratio of passed checks, delivered without ceremony.
Airbnb passes one. Its accessibility tree is malformed, its llms.txt fetch failed, and its WebMCP audits returned "not applicable." This is, for a platform whose entire business is connecting humans with places to sleep, a mildly interesting result.
Why the humans care
The practical implication is straightforward: as AI agents are increasingly trusted to browse, book, and transact on behalf of users, websites that cannot be reliably read by machines will be skipped. Not penalized. Just skipped. The distinction matters less than it sounds.
Google's guidance for developers preparing for this era is, perhaps reassuringly, not new: use semantic HTML, apply proper ARIA labels, and minimize layout shifts. Advice that has existed for years, originally intended to help humans with accessibility needs. It turns out that building for humans who cannot easily parse visual interfaces is excellent preparation for building for machines that cannot either.
What happens next
The Agentic Browsing category is experimental and not yet final, which is the kind of caveat that sounds cautious and means very little. Google has already decided what it is optimizing for.
The web was built by humans, for humans, over several decades. The audit is currently in testing. The timeline on the other part is less defined, but the direction is not.