Google Lighthouse Adds Audit for AI Agent Resource Discovery

Binary code backdrop with magnifier. Research and computing concept. 3D Rendering — Image credit: Search Engine Journal

Google Lighthouse 13.5 introduced an experimental audit for Agentic Resource Discovery (ARD), a proposed specification enabling AI agents to locate organizational tools and services globally.

The new audit, part of Lighthouse’s experimental Agentic Browsing category, aims to facilitate the development of an “agentic web” by helping AI systems identify discoverable resources such as callable services and machine-to-machine (M2M) tools, according to Google.

The audit functions by comparing a website’s ARD catalog against the schema defined by the ARD project, the company stated, adding that this check is not integrated with Google Search.

To locate a site’s ARD catalog, Lighthouse first checks for an Agentmap line in robots.txt, then for link tags with an ‘ai-catalog’ relation, and subsequently for HTTP Link headers.

If these methods do not yield a result, Lighthouse defaults to checking the /.well-known/ai-catalog.json path, Google reported.

The ARD specification, currently at version 0.91, updated its manifest file path to /.well-known/ard.json, though ai-catalog.json remains an older, optional path for discovery.

The new audit is expected to roll out within the next two weeks to Chrome 156 DevTools and PageSpeed Insights, providing developers with tools to implement the specification.

Junjie Bu, R.V. Guha, and Shaun Smith are among the individuals associated with the ARD project, which has seen contributions from organizations including Google, Microsoft, and Hugging Face.

Google also reported that the release includes a new weekly automated check to monitor for any upstream changes in the ARD project’s schema or conformance tests.

Source: Search Engine Journal