Confirmed
Lighthouse 13.5 Ships an Audit for AI Agent Resource Catalogs
Lighthouse 13.5, released September 18, added an 'ard-schema' audit under a new 'Agent Discoverability' heading. It validates Agentic Resource Discovery (ARD), a proposed spec for how AI agents find an organization's tools and services: check robots.txt for an Agentmap directive, then a link tag with the ai-catalog relation, then an HTTP Link header, defaulting to /.well-known/ai-catalog.json. It fails only on a schema error or an unloadable file; a missing catalog scores Not Applicable, and PageSpeed Insights follows within two weeks.
In plain termsThe ai-catalog.json file this audit checks for is a JSON manifest, a structured text file, listing what an AI agent can do on a site, like a menu built for machines instead of people; Lighthouse is the audit tool already built into Chrome that scores a page's speed and accessibility.
Why it mattersA site's agent catalog now feeds a mainstream audit tool millions of developers already run, so a missing or malformed one becomes a visible score hit instead of an invisible gap.
Our takeI found the same gap by hand in "I passed every agent-readiness check while three statements in my own metadata were false": a check that verifies a catalog exists and parses says nothing about whether its claims are true. Lighthouse's new audit is that same shape check running at far larger scale, and it will reward a schema-valid catalog full of stale or invented values exactly as it rewards an honest one.
Deep dive: What agent-readiness scanners actually check
Observed
Spotted: Public Gemini Notebook Pages Are Being Indexed as Spam on Google
SEO practitioners including Gagan Ghotra and Charles Floate flagged that publicly shared Gemini Notebook pages, the renamed NotebookLM, are being mass-created to publish spam that Google is indexing, covering coupon codes, peptides, and pornography. Search Engine Roundtable's Barry Schwartz cited a site-search count of more than 12,000 indexed pages. Google has not commented on the reports or said whether it will restrict public sharing or add spam controls to the feature.
In plain termsGemini Notebook is Google's AI research tool, formerly called NotebookLM, that can turn a set of uploaded documents into a shareable public page; 'indexed' means Google's search engine has crawled and stored that page so it can appear in results.
Why it mattersAny Google-owned surface that lets a logged-in user publish a public page becomes a fresh spam vector the moment it starts ranking, and the burden of noticing falls on outside researchers, not Google.
Our takeThis is the same failure mode as every prior Google feature that let anyone publish a public URL: spam fills the gap faster than moderation catches it. Watch whether Google throttles public sharing or de-indexes the pages before this becomes a larger cleanup story.
Observed
Spotted: Google Tests AI Overview Links That Route Into AI Mode Instead of the Cited Page
Marketer Gagan Ghotra posted video evidence on X showing Google testing anchor-style links inside AI Overviews that, instead of opening the cited web page, open a follow-up query inside AI Mode, Google's conversational search experience. Search Engine Roundtable's Barry Schwartz covered the find and framed it as another push from ordinary search results toward AI Mode. Google has not confirmed the test, described a rollout, or said which queries are affected.
In plain termsAI Mode is Google's separate, chat-style search experience; the concern here is that a link styled like an ordinary citation inside an AI Overview opens more chat instead of the website it appears to point to.
Why it mattersIf Google ships this, a link inside an AI Overview that looks like a source citation instead routes the click into Google's own AI Mode, one more click that never reaches the cited site.
Our takeThis would extend a pattern already visible in AI Mode itself, where a follow-up prompt sits one tap away and a citation click is not the path of least resistance. The number worth watching if Google confirms this is whether AI Overview click-through to sites drops in the same window.
Confirmed
Google Merchant Center Documents New Eligibility Thresholds for Store Ratings
Google's Merchant Center help page for Store ratings now spells out eligibility math that was previously vaguer: a store generally needs at least 100 eligible, unique reviews and an average of 3.5 out of 5 stars to qualify for paid text and Shopping ads, calculated from a rolling 24-month review window. Ratings are geo-segmented, scored separately per country rather than pooled globally, so a merchant can qualify in one market and not another.
In plain termsA 'rolling 24-month window' means Google only counts reviews from the last two years counting back from today, so old feedback ages out automatically as new reviews come in, rather than counting every review a store has ever received.
Why it mattersA merchant chasing a store rating badge needs to clear that 100-review, 3.5-star bar inside each country's own rolling 24-month window, not as one global total, which changes where review-generation effort should go first.
Our takeA documented threshold is a compliance target, not a new bar Google invented today, since most merchants already near or past 100 reviews were effectively meeting it already. The thing to check next is whether the same reorganized help page changed anything about how an ineligible country's badge displays, since the geo-segmentation rule is the easiest part to miss.