60-Day Tests Confirm Top LLMs Ignore Open Knowledge Format Mirrors and Default to HTML
Google released Open Knowledge Format (OKF) v0.1 as a vendor-neutral standard for enterprise agents parsing internal databases and APIs.
Practitioners hijacked the spec for public-site generative optimization, mirroring standard pages into /okf/ folders containing raw .md files mapped by an index.md.
Each markdown file requires YAML frontmatter injecting the type, title, description, standard resource URL, and specific tags.
A 60-day field test of this architecture utilizing /audits/<slug>.md mirrors alongside standard HTML pages exposes current crawler behavior.
Google Search Console crawls the .md URLs within a few days, though standard search impressions remain essentially zero.
Crucially, ChatGPT and Perplexity crawlers ignore the markdown corpus entirely.
Throughout the tracking window, LLMs consistently bypassed the OKF folder, cited the standard HTML URLs exclusively, and triggered zero change in overall citation velocity.
The architecture fails as a direct ranking injection but operates as frictionless content-distribution infrastructure.
Practitioners deploy the OKF folder alongside llms.txt—mapping YAML-typed frontmatter for knowledge graphs while llms.txt maps standard links—to feed offline pipelines.
Developers and researchers pull these raw .md mirrors directly into Claude Code sessions and offline tools.
The setup costs almost nothing to generate and positions the corpus for future agentic ingestion without relying on immediate search visibility.