Google Lighthouse 13.5 is expanding beyond classic page performance and SEO checks. The latest release adds an experimental audit for Agentic Resource Discovery, or ARD, a proposed standard for how AI agents find the tools and services an organization can offer.
It’s another sign that machine-readable discoverability is becoming an early, if still fuzzy, marketing channel.
What the new audit actually checks
The audit validates a site’s catalog against the ARD project’s schema. It hunts for a catalog in a few places, in this order:
- An Agentmap line in the robots.txt file.
- A link tag with an ai-catalog relationship.
- An HTTP Link header using the same relationship.
- If none of those exist, the fallback path is /.well-known/ai-catalog.json.
If nothing is found, Lighthouse marks the audit as Not Applicable. Schema errors and pointers to unloadable catalogs fail. It’s not scored from 0 to 100; the experimental Agentic Browsing category reports a pass ratio instead.
Why “Not Applicable” can feel confusing
Here’s the catch: the ARD spec has since moved its preferred manifest location to /.well-known/ard.json. The older ai-catalog.json path remains optional, but the Lighthouse 13.5 source appears to still look for the older names and does not yet check for ard.json or the ard link relation as of the release.
That means a modern site using the current preferred file could show Not Applicable because Lighthouse didn’t find a catalog in its current search areas—not because the site lacks an ARD manifest. The audit is also separate from Google Search. It doesn’t mean your organic rankings are at risk.
ARD vs. llms.txt and WebMCP
The ARD proposal is about more than content summaries. It can help AI systems find MCP tools, A2A agents, skills, and other callable services through search. The specification is still a proposal, with authors from Google, Microsoft, and Hugging Face.
By comparison, llms.txt is a plain content summary for agents, and WebMCP gives a page structured actions an agent can call once it is already there. ARD is the discovery layer that points to tools and services across the web. They are complementary, not interchangeable.
Put another way: llms.txt answers “what is this site about?” WebMCP answers “what can this page do?” ARD answers “what can this organization do for an agent?” That framing helps teams avoid creating overlapping files.
What to do right now
Don’t chase a Lighthouse pass just yet. The standards are emerging, and the audit is expected to hit Chrome 156 DevTools and PageSpeed Insights within two weeks. Still, forward-looking D2C and SaaS teams can start mapping their public services, APIs, booking flows, and commerce endpoints into a structure an agent could parse. Assign one owner to keep that inventory current, and revisit the ARD spec as it stabilizes. The brands that document their offerings clearly now will have a head start if agentic discovery becomes a real traffic source.
Source: Search Engine Journal
