Google Lighthouse Adds Audit For AI Agent Resource Discovery

Google’s Lighthouse 13.5 introduces an audit for Agentic Useful resource Discovery (ARD), which is a proposed spec for the way AI brokers find the instruments and providers provided by a company. The audit compares a web site’s ARD catalog with the schema outlined by the ARD challenge. If the location doesn’t present a catalog pointer, Lighthouse defaults to /.well-known/ai-catalog.json. The newest spec, dated Aug. 26, moved this file to ard.json, with the outdated filename remaining optionally available for software program that reads it.

The Lighthouse crew expects this update to roll out in Chrome 156 DevTools, and in PageSpeed Insights inside two weeks.

This audit is a part of Lighthouse’s experimental Agentic Shopping class, separate from web optimization audits, and it’s not tied to Google Search wherever within the release notes. The studies will present an ARD end result or Not Relevant if no catalog is discovered.

What The Audit Checks

Within the 13.5 source code, Lighthouse checks the robots.txt file for an Agentmap line, which is a directive from the ARD specification pointing to a catalog. It then seems to be for a hyperlink tag with the ai-catalog relation and likewise checks the HTTP Hyperlink header for a similar relation. If not one of the three is on the market, it requests /.well-known/ai-catalog.json.

If there isn’t a pointer and the well-known path doesn’t serve the file, the audit is marked as Not Applicable. Schema errors trigger failure, as does a pointer to an unloadable catalog. The pull request describes ard-schema as a “schema conformance audit.”

The identical launch groups the new audit with the existing llms.txt audit underneath a heading referred to as Agent Discoverability. Google’s docs say the Agentic Shopping class doesn’t produce a 0-100 rating and exhibits a go ratio as a substitute, as a result of “the requirements for the agentic net are nonetheless rising.”

The Spec Has Since Renamed The File

ARD v0.91 now specifies /.well-known/ard.json because the manifest location and mentions ai-catalog.json as the sooner path. When I covered ARD’s launch in June, the spec was a v0.9 draft, with ai-catalog.json because the file for organizations to publish.

In v0.91, software program studying these information is required to fetch ard.json and also can examine the older path. The spec cautions {that a} file solely accessible on the outdated path “will not be discovered.”

I didn’t discover any references to ard.json or the ard hyperlink relation within the 13.5 supply or in the primary department of the challenge as of Sept. 21. The spec additionally mentions discovery by in-page JSON-LD and DNS data, however the 13.5 code doesn’t examine both. Lighthouse built its validator based on the ARD project’s own conformance tests.

The specification remains to be a proposal and credit three authors: Junjie Bu from Google, R.V. Guha from Microsoft, and Shaun Smith from Hugging Face.

How ARD Differs From llms.txt And WebMCP

ARD is all about discovering sources. It consists of MCP instruments, A2A brokers, expertise, and different callable providers, so AI methods can discover them by search.

An llms.txt file summarizes your web site’s content material for brokers. WebMCP, which I shared about in August, permits a web page to supply structured actions that an agent can name as soon as it’s in your web site.

Why This Issues

Primarily based on Lighthouse 13.5 supply code, if a web site makes use of /.well-known/ard.json with rel="ard", doesn’t have an Agentmap line, and doesn’t use the older ai-catalog names, it might be labeled as Not Relevant.

This merely means Lighthouse didn’t discover a catalog in its present search areas. However it doesn’t essentially imply the location is lacking a present ARD manifest.

As I shared in May, when Lighthouse launched its llms.txt audit, do not forget that Google Search’s steering and Lighthouse’s agent checks are various things.

Wanting Forward

Lighthouse expects 13.5 to achieve PageSpeed Insights inside two weeks of the discharge.

The discharge additionally provides a weekly automated check that flags when the ARD challenge’s schema or conformance take a look at modifications upstream. The locations Lighthouse seems to be for a catalog are set individually, within the code that fetches it.

Featured Picture: Golden Dayz/Shutterstock