Google Lighthouse 13.5 Adds Audit for AI Agent Resource Discovery

Google's Lighthouse 13.5 introduces a new audit for Agentic Resource Discovery (ARD), a proposed specification designed to help AI agents locate tools and services offered by organizations online. The update is expected to roll out in Chrome 156 DevTools and PageSpeed Insights within two weeks of its September 21, 2026 release.

As AI agents become more deeply embedded in how people search and interact with the web, the question of how those agents find and verify available tools is growing in urgency. This audit signals a broader industry push to standardize how websites communicate their capabilities to autonomous systems — a shift that could meaningfully reshape technical SEO practices and web development workflows in the months ahead.

---

What Lighthouse 13.5 Actually Checks

The new audit validates a website's ARD catalog against the schema defined by the ARD project. When no catalog pointer exists, Lighthouse defaults to checking /.well-known/ai-catalog.json. The audit sits within Lighthouse's experimental Agentic Browsing category and is separate from existing SEO audits. Based on available release notes, it carries no direct connection to Google Search rankings.

In practical terms, the 13.5 source code reveals that Lighthouse runs through a sequence of checks to locate a valid catalog:

• It scans robots.txt for an Agentmap line as directed by the ARD specification
• It looks for a link tag carrying the ai-catalog relation
• It checks the HTTP Link header for that same relation
• If none of the above are present, it requests /.well-known/ai-catalog.json

If no pointer is found and the well-known path does not serve the file, the audit returns a Not Applicable result rather than a failure. Schema errors or a pointer leading to an unloadable catalog will trigger a failure. The pull request describes the audit as a "schema conformance audit."

How Results Are Displayed

Google's documentation notes that the Agentic Browsing category does not produce a traditional 0–100 score. Instead, it displays a pass ratio — because, as Google puts it, "the standards for the agentic web are still emerging." The same release groups the new ARD audit alongside the existing llms.txt audit under a shared heading called Agent Discoverability.

For teams already familiar with Google's tools for improving website performance and visibility, the addition of a dedicated agentic browsing category represents a meaningful expansion of what Lighthouse evaluates — moving beyond human-facing performance into machine-facing discoverability.

---

The Spec Has Evolved — and Lighthouse Hasn't Fully Caught Up

The ARD specification has moved quickly since its initial draft. When ARD v0.9 launched in June 2026, ai-catalog.json was the designated file for organizations to publish. By August 26, 2026, ARD v0.91 updated the manifest location to /.well-known/ard.json, with the older ai-catalog.json path remaining optional for backward compatibility.

Software reading these files is now required to fetch ard.json first and may also check the older path as a fallback. The spec cautions that a file available only at the old path "may not be found."

Where the Current Implementation Falls Short

As of September 21, the 13.5 source code and the main project branch showed no references to ard.json or the ard link relation. The spec also describes discovery through in-page JSON-LD and DNS records, but neither is checked in the current Lighthouse implementation.

This gap has a direct practical consequence: a website using /.well-known/ard.json with rel="ard" — and none of the older naming conventions — would currently receive a Not Applicable classification, even if a valid ARD manifest exists. For teams building to the latest version of the spec, this is a material risk worth planning around now.

The ARD specification credits three co-authors: Junjie Bu from Google, R.V. Guha from Microsoft, and Shaun Smith from Hugging Face.

Lighthouse has also confirmed that 13.5 adds a weekly automated check that flags changes to the ARD project's schema or conformance tests upstream — a clear signal that the team anticipates the spec will continue evolving at pace.

---

How ARD Fits Into the Broader Agentic Web

A Distinct Role Among Emerging Standards

ARD is one of several emerging standards designed to help AI agents navigate and interact with websites more effectively — and it occupies a distinct role compared to its peers. An llms.txt file summarizes site content for agents. WebMCP, introduced in August 2026, allows a page to offer structured actions an agent can call once it arrives on the site. ARD focuses specifically on discovery — surfacing MCP tools, A2A agents, skills, and other callable services so AI systems can find them before ever landing on a page.

The analogy to XML sitemaps is instructive: a behind-the-scenes infrastructure layer that most users never see but that profoundly shapes how machines understand and interact with the web. The organizations that invested early in sitemaps gained a quiet but durable advantage in how search engines crawled and indexed their content. ARD may follow a similar trajectory.

Understanding how these developments connect to real-world artificial intelligence adoption across business sectors is increasingly relevant for any team making infrastructure decisions today.

What This Means for Web Developers and SEO Professionals

The practical implications break down into three clear areas of action.

First, teams should audit their current implementation against both ai-catalog.json and the newer ard.json path. Relying solely on the older naming convention risks being inadvertently classified as Not Applicable once Lighthouse updates its lookup logic to reflect the v0.91 specification.

Second, given that the Agentic Browsing category sits entirely outside Google Search audits, ARD compliance is best treated as forward-looking infrastructure investment rather than an immediate ranking factor. The groundwork laid now will matter significantly more as agent-driven discovery matures.

Third, staying current with ARD specification versioning is not optional — the gap between v0.91 and Lighthouse's current implementation demonstrates how quickly these standards move and how costly it can be to build against an outdated draft. Subscribing to the ARD project's release notes and monitoring the official Lighthouse changelog are straightforward ways to stay ahead of breaking changes.