Lighthouse 13.5 ships an audit for AI agent resource discovery, but its checks don't match the current ARD proposal. Here's what that gap means for sites betting on agent traffic.
Google’s Lighthouse tool now includes an audit aimed at AI agent resource discovery, shipping in version 13.5. The catch, reported by searchenginejournal.com on September 21, 2026, is that Lighthouse’s checks diverge from the latest ARD proposal. Two things claiming to describe the same thing, measuring it differently, is the whole story here.
What shipped
Lighthouse 13.5 adds an audit category for AI agent resource discovery. In plain terms, that means the tooling now has an opinion about whether an AI agent can find and use the resources a site publishes for it. That’s a new axis for a tool that has spent its life grading page speed, accessibility, and SEO fundamentals.
The audit’s checks differ from the most recent ARD proposal. searchenginejournal.com’s report does not spell out which specific checks diverge or how far apart the two are, so treat the size of the gap as an open question rather than a known quantity. What’s established is that the divergence exists.
Why a Lighthouse audit matters more than a spec
Specs get written by people who care about specs. Lighthouse gets run by people who care about shipping. That asymmetry is why this is more than a footnote in a changelog.
For the last several years, “optimize for AI” has been a consulting phrase with no score attached. You could argue about llms.txt, structured data, and crawl directives all day and never produce a number anyone could act on. A Lighthouse audit changes that. It puts agent resource discovery into the same pass/fail rhythm as everything else in the report, which means it will get triaged, ticketed, and assigned to someone. That’s how a concept becomes a work item.
The risk is equally obvious. When a major tool ships a check before a spec settles, the check becomes the de facto standard whether or not it’s the right one. Developers optimize against the audit because the audit is what they can see and run. If Lighthouse and ARD disagree on the details, sites will end up building for Lighthouse, and the ARD proposal will have to either converge or lose the argument by default.
The divergence is the interesting part
Two possibilities, and they lead to different responses.
The first is that Lighthouse is ahead. Google’s tooling teams often ship against internal signals before public proposals catch up, and an audit that reflects what Google’s own agents actually look for would be more useful than one that reflects a committee draft. If that’s the case, the ARD proposal is the thing that needs to move.
The second is that Lighthouse is behind, or checking a narrower slice of what ARD describes. That happens too, and it’s the more common failure mode for early audits: they test the easy, mechanical part of a concept and leave the hard part unmeasured. A site could pass the audit cleanly and still be invisible to agents that follow the fuller ARD spec.
Nothing in the reporting settles which one this is. Anyone telling you they know is guessing.
What to actually do with this
Run the audit. It’s free, it’s in a tool you already have, and it will tell you something about how your site presents resources to automated consumers. But don’t treat a green score as proof that agents can find what you’ve published for them.
The practical posture is to keep the two definitions separate in your head. Lighthouse 13.5 tells you whether you pass Lighthouse 13.5. The ARD proposal tells you what a group of people currently think agent resource discovery should look like. Neither one is the ground truth, because there isn’t one yet.
If you’re building for agent traffic in any serious way, the useful move is to track both and note where they disagree. Those disagreement points are where you’ll find the work that actually matters, and they’re also where the standard is most likely to shift under you in the next few releases.
What to watch next
Three things would tell you how this resolves.
- Whether Lighthouse’s next release narrows the gap with ARD, or whether the ARD proposal changes to match Lighthouse. Movement in either direction is a signal about who’s setting the terms.
- Whether the audit’s checks get documented in enough detail that developers can reason about them, rather than just reacting to a score.
- Whether other agent-facing tooling adopts the same checks. A single audit is a data point; three tools agreeing is a standard.
For now, the honest read is that agent resource discovery has graduated from a topic people argue about on conference panels to something with a number attached. That’s progress. It’s also the moment when a lot of sites will start optimizing for a definition that may not survive contact with the next version of the spec.
Source: searchenginejournal.com