Content — Serving Machine-Readable Pages to Agents

The agent-readiness dimension that serves machine-readable, answer-first pages — markdown twins, content negotiation, JSON-LD, meta and canonicals.

What does the content dimension mean?

The content dimension makes the substance of a page readable by machines, not only rendered for humans: the same meaning served as a markdown twin, negotiated by media type, structured as JSON-LD and written answer-first, so the first sentence of every section is the sentence an agent can lift. Aggarwal et al. (2023) recorded up to 40 % higher visibility in generative-engine responses for pages optimized for those engines.

Which signals and standards does the content dimension cover?

Five signals, each tied to a spec an agent can check:

• Markdown twins — a .md copy of every HTML page, so an agent reads the text without parsing layout markup, scripts or navigation.
• Content negotiation — RFC 9110 (IETF, 2022) lets the server pick the representation from the Accept header and declare that choice with Vary; the media type text/markdown is registered by RFC 7763 (IETF, 2016) and requires a charset parameter. One URL then serves humans HTML and agents markdown.
• JSON-LD / schema.org — JSON-LD 1.1 is a W3C Recommendation (W3C, 2020) and the format Google recommends for structured data (Google, 2025), so an agent extracts facts as data instead of re-parsing prose.
• Clean meta, canonical and Open Graph — an accurate title and description, one canonical URL per page under the canonical link relation (IETF, 2012), and og:title plus og:image for consistent representation wherever the page is referenced.
• Answer-first structure and FAQ — each section opens with a standalone answer, and visible question-answer pairs are mirrored word-for-word in FAQPage JSON-LD. Google stopped showing FAQ rich results in Search on 7 May 2026 (Google Search Central, 2026); for agents the value is the pre-structured pair itself, not a search feature.

How does the Agent-Readiness Audit score content?

Content is the Audit's largest category: 7 of 18 checks and 39 of 100 points — markdown-negotiation (9), json-ld (7), markdown-alternate (5), vary-accept (5), meta-description (5), canonical (4) and open-graph (4) (Agents Welcome, 2026). Each rule is fetch-and-inspect: a request with Accept: text/markdown must return Content-Type: text/markdown; the HTML must carry <link rel="alternate" type="text/markdown">, a Vary header naming Accept, a meta description of at least 20 characters, a canonical link, og:title with og:image, and a JSON-LD block that parses. This site passes all seven live: every page has a .md twin, and requesting this URL with Accept: text/markdown returns markdown with Vary: Accept.

Content — frequently asked questions

What is a markdown twin?

A markdown copy of an HTML page at the same URL, reached by appending .md or by sending Accept: text/markdown, so an agent reads the text without layout markup. Every page on this site has one.

Which media type should a markdown twin use?

text/markdown, registered by RFC 7763 with a required charset parameter, for example Content-Type: text/markdown; charset=utf-8. The response should also carry Vary: Accept so caches keep the HTML and markdown versions apart.

Does FAQ schema still matter now that Google dropped FAQ rich results?

For agents, yes. A visible question-answer pair mirrored word-for-word in FAQPage JSON-LD is a pre-structured chunk an answer engine can lift. Google stopped showing FAQ rich results in Search on 7 May 2026, so treat FAQPage as an agent signal, not a search feature.

Related: schema.org for agents · serve markdown twins · return text/markdown · get cited by AI (GEO) · audit your site

← Agent-Readiness · .md