Google Lighthouse 13.5 adds an Agentic Resource Discovery audit, which checks whether a site publishes a machine-readable catalog for AI agents

Google Chrome released Lighthouse 13.5 on 2026-09-18 with a new audit for Agentic Resource Discovery, an emerging web standard for a machine-readable catalog that tells AI agents where a site's resources live. The release notes list the audit under New Audits as ard-schema: add Agent Resource Discovery audit, and separately note that "llms.txt and ARD audits" are now grouped under a heading called "agent discovery". Search Engine Journal reported the ship on the same day, and its news writer Matt G. Southern first noticed one difference between the audit and the current specification, described below.

What Lighthouse checks

The audit looks for a JSON catalog and compares it against the schema published by the ARD project. According to Search Engine Journal, it defaults to /.well-known/ai-catalog.json when the site does not give a pointer, and it also reads robots.txt for an Agentmap directive, HTML for a link tag with the ai-catalog relation, and HTTP Link headers pointing at a catalog. The audit then scores the catalog against the schema and reports any fields that do not fit.

Two clean-up entries in the release notes confirm the shape: a fixture at .well-known/ai-catalog.json was checked in with the release, and a smoke test was added that runs the audit end to end.

Where the spec has already moved on

The ARD project shipped v0.9 in June and updated to v0.91 on 2026-08-26, according to the Search Engine Journal write-up. The newer version specifies /.well-known/ard.json as the default location, with the older /.well-known/ai-catalog.json allowed for compatibility. Lighthouse 13.5 still checks the older path. The spec also describes four discovery methods, DNS records and in-page JSON-LD included, and Lighthouse currently only looks at well-known files. Both differences will matter to anyone hand-writing the catalog this week.

When teams will see it

The release notes and the Search Engine Journal report both say the audit ships in Chrome DevTools 156 and will reach PageSpeed Insights within two weeks. Anyone already running Lighthouse from a CI pipeline can pull v13.5.0 today and get the new audit in the same run as its Core Web Vitals and llms.txt scores.

For a site that already publishes llms.txt, adding an ai-catalog.json is now a small, well-scoped follow-up: the catalog is a JSON file, and Lighthouse tells you whether it fits the schema before an agent ever reads it. The two remaining questions, whether to write the file at ai-catalog.json for the current audit or at ard.json for the current spec, and whether the file changes what AI agents actually do given that Googlebot itself uses different signals, are the ones a team should decide before publishing.

Source

• Google Chrome: Lighthouse v13.5.0 release notes on GitHub
• Search Engine Journal: Google Lighthouse Adds Audit For AI Agent Resource Discovery