Serving Humans and AI Through Content Negotiation (www.nibzard.com)

The author rolled out a dual-format content delivery system that serves the same underlying content as rich HTML for humans (/{slug}) and as clean, metadata-rich Markdown for machines (/{slug}.md or when Accept: text/markdown is preferred). Rather than cloaking or restricting endpoints, a middleware layer inspects URL patterns and Accept headers to pick the best representation, enforces content parity (no hidden differences), and excludes drafts at query time. Key deployment details: content is authored as markdown files with extensive metadata (title, tldr, date, tags, author), canonical links point to the HTML page for SEO, and the stack uses Cache-Control plus Vary: Accept for performant negotiation. The repo exposes structured artefacts and endpoints (llms.txt, rss.xml, /api/raw/[slug], /api/og/[slug]) and follows standards like RFC 7231, Schema.org, RSS 2.0 and llms.txt. For the AI/ML community this matters because it gives agents clean, parseable inputs with explicit attribution and citation-ready headers—reducing parsing errors and improving provenance—while preserving a rich human UX. The approach is portable to static site generators, supports discoverability via permissive robots and sitemap, and future-proofs content delivery for diverse consumers (voice, AR/VR, research tools). In short: same content, two optimized presentations—machine-friendly structure and metadata for models; interactive, accessible pages for people.