The AI search material was the loosest part of the skill. It named the terms without saying what they meant, and it made recommendations without saying how well any of them were established. This release fixes both.

Added

Working definitions for GEO, LLMO and AEO, each tagged with an evidence level: High, Medium or Experimental. A recommendation now carries its own confidence.
An eight-stage generative visibility audit with six-field findings, covering entity clarity, original information gain, and citation-ready editorial checks.
A repeatable measurement method for AI referrals, citation and mention rate, and source prominence — so two runs a month apart are comparable.

Changed

Training, search and user-triggered retrieval are now classified separately. Blocking a training crawler and blocking a search crawler are different decisions with different costs, and the previous text let them blur together.
Google-Extended’s mixed purpose and its blocking limitations are stated explicitly.
llms.txt guidance is marked optional and experimental. Google states it does not use it.
Documentation, research provenance and both distribution manifests are synchronised at 1.5.0.

Fixed

edge-seo.md — documented the navigation-request rule. With a Worker script and not_found_handling set, a browser navigation that misses a static asset is served 404.html without invoking the Worker, while crawlers still reach it. That splits the site in half: Googlebot indexes the SSR page, and every human who clicks the result lands on a 404. The file’s own reference config had the bug. Three mitigations added, plus a browser-vs-curl check to detect it.
astro-seo.md — noted the Astro 7.2.0 Astro.url.pathname fix for build.format: 'preserve', and the @astrojs/sitemap 3.7.3 and 3.7.4 fixes.

Verified on 2026-09-07. Reference files that were not touched keep their earlier dates.