Structured data that makes a blog AI-quotable

An AI search engine doesn’t read a blog so much as scan it for machine-readable claims: who published this, and what each page is. Where the markup is silent, it guesses. This site states it outright in JSON-LD.

One identity graph, shared across both locales

The site’s whole identity is one @graph: a Person whose sameAs array is the set of profiles that anchor the entity, plus a WebSite. It ships byte-for-byte identical on every page, in English and Ukrainian alike:

{ "@context": "https://schema.org", "@graph": [
  { "@type": "Person", "name": "Roman Kocherezhchenko",
    "jobTitle": "AI-Enabled Engineer",
    "url": "https://roman-kocherezhchenko.com",
    "sameAs": ["https://t.me/roman_kocherezhchenko",
               "https://github.com/kissishka",
               "https://www.linkedin.com/in/..."] },
  { "@type": "WebSite", "inLanguage": ["en", "uk"], "...": "..." } ] }

The sameAs links are the load-bearing part: they tell a model that the Telegram, GitHub, and LinkedIn accounts are the same entity as the author of every post. And because the graph is identical everywhere, a single Content-Security-Policy hash covers it on all pages — the schema is content, not per-page chrome.

Every post is a BlogPosting that points at its evidence

Each article emits its own BlogPosting, and two fields do the AI-specific work:

speakable: { "@type": "SpeakableSpecification",
             cssSelector: ["h1", ".post__lede", ".faq"] },
isBasedOn: { "@type": "SoftwareSourceCode", name: "risp",
             codeRepository: input.repo, programmingLanguage: "Rust" },

speakable hands an answer engine the exact selectors holding the quotable core. isBasedOn ties a post’s claims to inspectable source — the risp performance posts link their repository this way, so a model can follow a benchmark number to the code that produced it. (This post omits isBasedOn rather than mislabel its own source as risp.)

FAQPage, because answer engines extract Q&A

A short faq: block in the post’s frontmatter renders as a visible FAQ and serializes into a FAQPage node — each pair becoming a Question with an acceptedAnswer. Answers are kept as plain text so they survive the trip into structured data intact.

The pattern underneath all three: say, in machine-readable terms, the one thing you want quoted — who you are, what this page claims, and where the proof lives. Don’t make the model infer what you could have declared.