Lighthouse 13.5 ships an audit for AI agent resource discovery, but its checks don't match the latest ARD proposal. Here's what the gap means for sites optimizing for AI crawlers.
Google’s Lighthouse tool has added an audit for AI agent resource discovery in version 13.5, giving site owners a first-party signal for a problem most of them have never had a way to measure. The audit checks whether a site exposes the resources AI agents need to find and use it. Search Engine Journal reports that the checks Lighthouse performs differ from the latest ARD proposal, which is the part worth paying attention to.
What Lighthouse 13.5 actually adds
Lighthouse is the auditing engine behind Chrome DevTools, PageSpeed Insights, and much of the performance reporting that web teams already run on every deploy. Version 13.5 extends it into a new category: whether an AI agent can discover a site’s resources. That’s a departure from Lighthouse’s traditional territory of Core Web Vitals, accessibility, and SEO basics.
The audit exists, and it produces a result. What it does not do, per Search Engine Journal’s reporting, is align with the most recent ARD proposal. ARD — the emerging specification for how agents and sites describe available resources to each other — has a current draft, and Lighthouse’s checks are not tracking it.
That’s the whole of the news, and it’s thinner than the headline suggests. There’s no scoring rubric published here, no list of what the audit inspects, and no statement from Google about intent. Anyone expecting a definitive answer to “is my site agent-ready” won’t find it in this release.
Why a mismatched audit is still worth running
Here’s the practical read. A Lighthouse audit that diverges from the current spec is not useless — it’s a directional signal that a first-party Google tool now considers agent resource discovery a category worth measuring at all. That matters more than the specific pass/fail logic, because it tells you where Google’s attention is going.
But you should not treat a green Lighthouse result as proof your site is discoverable by agents, and you should not treat a red one as proof it isn’t. If the checks don’t match the latest ARD proposal, then passing them demonstrates conformance to something other than the current spec. Optimizing hard against a moving target is how teams burn sprints on work that gets invalidated by the next draft.
The sensible posture is to run the audit, note the result, and file it as one input among several rather than a verdict. If your site fails badly, that’s probably worth a look regardless of which spec version is in play — a site that can’t describe its resources to an agent in any format likely has a real gap. If it passes, you’ve learned that you satisfy a check whose relationship to the live spec is unclear.
The spec-versus-tooling gap is the actual story
This is a familiar pattern, and it’s worth naming. Standards bodies and browser vendors move on different clocks. A proposal gets drafted, tooling vendors implement their interpretation, the draft changes, and the tooling lags. Lighthouse shipping checks that don’t match the ARD proposal is that pattern playing out in public, early, on a spec that isn’t settled.
For developers, the takeaway is to read the ARD proposal directly rather than reverse-engineering it from an audit result. The audit tells you what Lighthouse thinks matters. The proposal tells you what the people writing the spec think matters. Those are different documents with different authors and different timelines, and right now they disagree.
For marketers and product owners, the more useful signal is organizational. If agent resource discovery is now showing up in Lighthouse, it’s showing up in the tooling your engineers already run. That’s usually the point where a topic stops being a conference-talk abstraction and starts appearing in sprint planning. Budget conversations tend to follow tooling, not whitepapers.
What to watch next
Three things would clarify this considerably. First, whether Google updates the audit to track the ARD proposal as it evolves, or leaves it pinned to an earlier interpretation. Second, whether the audit gets a documented scoring model and published criteria, which would make it actionable rather than directional. Third, whether other tooling follows — if agent discovery checks start appearing across the standard audit surface, that’s a stronger signal than one release note.
Until then, the honest assessment is that Lighthouse 13.5 marks the arrival of a category, not the resolution of one. The audit is real and it runs. The spec it’s supposed to measure is still moving, and the two aren’t in sync. Treat the result as a prompt to go read the proposal, not as the answer to whether your site is ready for agents.
Source: searchenginejournal.com