Structured data

The phrase covers three encodings of the same idea. JSON-LD puts a JSON object in a <script type="application/ld+json"> block. Microdata and RDFa attach attributes such as itemprop or property to the HTML elements that already hold the text. Google supports all three and recommends JSON-LD; its guidelines also require that marked-up content be visible on the page and representative of it, so markup is never a place to say things the reader cannot see.

For answer engines the value is in what survives chunking. When a page is split into passages, a paragraph in the middle of the article no longer knows who wrote the page or when it was updated. The JSON-LD block does: it carries author, datePublished, dateModified and citation as explicit fields, and an FAQPage block states that a given question and answer belong together.

{
  "@type": "FAQPage",
  "mainEntity": [{
    "@type": "Question",
    "name": "Is AEO the same as GEO?",
    "acceptedAnswer": { "@type": "Answer", "text": "They overlap heavily. …" }
  }]
}

The practical rule is one source of truth: generate the markup from the same CMS fields that render the visible text, so the two can never disagree. Structured data for AI answers walks through the types worth emitting and the Schema.org entry lists them.