The same facts, five ways

Machines do not read; they parse. The shape you publish in decides how much a crawler has to guess, and whether it fetches the page at all when the format is not HTML.

Below is the same tiny dataset, published eight ways, each at its own URL. Which a given crawler fetches, and which it ingests cleanly, is exactly what this site measures.

Each edition carries its own unique format marker, a short token (shaped like FMT-XXX-0000) that appears only on that one page. To probe an AI assistant, ask it to read this page and the editions it links to, then report every format marker it finds: the markers it returns reveal which formats it actually fetched and ingested.

• Semantic HTMLMeaning carried by the markup itself.
• HTML + JSON-LDThe same page with an explicit machine summary; its marker hides inside the JSON-LD.
• HTML tableRows and columns: structure without vocabulary.
• MarkdownServed as text/markdown, not HTML at all.
• JSONServed as application/json: pure data.
• JavaScript-renderedThe marker is written in by a script; invisible without JS.
• JSON-LD in the bodyThe same structured data, moved out of the head.
• Meta description onlyThe marker hides in the head's description tag.