Structured data: an explicit reading

Prose is ambiguous; machines guess. Structured data removes the guessing by stating, in a fixed vocabulary, what a page is about.

The common form is schema.org vocabulary embedded as JSON-LD — a small block of JSON in the page head that names the type of thing (an Article, a Product, a Recipe), its title, author, and key fields. A crawler that would otherwise infer all of this from the text gets it handed over, unambiguously.

This very page carries an Article block in its head. It duplicates what the prose says, but in a shape a machine can trust without natural-language understanding. For AI agents that assemble answers from many sources, that trustable shape is often the difference between being cited correctly and being paraphrased wrong.

Formats are a spectrum: clean semantic HTML, JSON-LD, plain Markdown, raw tables. Each is easier or harder for a given reader. The safe move is to say the important things in more than one of them.