Context
The site had a BreadcrumbList on every page since ADR-0021 and one hand-written WebSite and Person block on the home page; no page said in markup who wrote it, and the canonical definitions on Terms and in clause 3 of the definition draft were plain text. Stefan Coetzee wants these pages and terms to be found and attributed to him, on the honest side: every claim in the markup true and visible on the page.
Google's guidance on its AI features (developers.google.com/search/docs/fundamentals/ai-optimization-guide) says structured data is not required for them and that no special schema exists for them; it names indexable, snippet-eligible, useful pages, a good page experience and content that can be crawled without JavaScript. Google ignores llms.txt. Schema.org markup still states the author, the page type, the dates and the meaning of a term in a form that search engines and other tools can read, at little cost.
Decision
page_ld() in scripts/site_chrome.py writes one JSON-LD graph in the head block of every page, next to the BreadcrumbList:
- Person: Stefan Coetzee, with the profiles in
site.personofsite/nav.yml(LinkedIn, Substack, GitHub, this site, X) and the job title the home page carried. One node, one@id, referenced as author and publisher of every page. - WebSite: the site, with the site search as a
SearchAction(/inside/search/?q=). - The page:
Articlefor research pages,TechArticlefor docs pages,CollectionPagefor the section and topic hubs,WebPagefor the rest; headline, description,datePublished,dateModified, keywords, the breadcrumb by@id. Dates come from the page: docs front matter (created, updated), thepublisheddate a research page gives itself in its header (recorded insite/nav.yml), and thesitemap.xmldate as the last change. A page without a publication date of its own gets none. Two pages name a model as their writer or scorer; their markup carries it ascontributor. - DefinedTermSet: on Terms one DefinedTerm per entry and on the definition draft one per entry of clause 3, each with the wording on the page, read from the page at build time, so the markup cannot drift from the text.
The home page's hand-written block went into the shared Person; its subreddit links stayed out, because a community is not a profile of the person. No FAQPage markup: no page is questions and answers. site_chrome.py parses every JSON-LD block on every page and reports invalid JSON as an error, which fails --check.
Consequences
A new page gets the markup on the next run of the chrome. A research page that states a publication date needs it in site/nav.yml as published. The legislation-track builders run the chrome after writing, so the definition draft's terms follow its text. The markup is not expected to bring citations in AI answers by itself; what Google names for that is useful, crawlable pages. So the pages that fill their lists in the browser (Inside, the board, Mission Control) also carry the same lists as static HTML, written at build time by scripts/build_fallbacks.py, and no page carries a noindex except the 404 page. Whether any of this changes citations is not measured here; the measurement needs search-console accounts and is a separate decision.