Content for AI retrieval
Make content quotable, citable, and retrievable by AI systems without losing precision.
Updated 2026-09-29
Patterns that improve retrieval quality:
- Stable fragments: Use clear headings and anchors. Keep definitions and spec tables tight and self‑contained.
- Parallel formats: Offer HTML for reading, plus CSV/JSON for ingestion. Match IDs across formats.
- Tables that parse: One header row, unit columns, and cell values free of inline units where possible.
- Canonical wording: Prefer standard terms and symbols; add synonyms in a visible glossary or metadata.
- Passage length: 40–120 words per reference paragraph is easier to cite verbatim.
Anti‑patterns:
- Specs buried in images/PDF only.
- Marketing copy where a definition or constraint is expected.
- JS‑only rendering that hides content from fetchers.