Glossary Updates6 new terms added to the glossaries · October 11, 2026, 13:38 CEST
AI TechDocKnowledge

Glossary · 5 · Content lifecycle and AI

Curated corpus

Also known as: Curated knowledge base, Curated retrieval corpus, Approved source collection

In AI functions that answer from documentation, a curated corpus is a controlled collection of released, reviewed content that the function may retrieve from or be tuned on, with defined rules for what enters and leaves it, as opposed to an uncurated collection of whatever content is available.

  • Content lifecycle
  • AI
  • Technical documentation

In one sentence

A curated corpus is a controlled collection of released content an AI function may retrieve from or be tuned on, with rules for what enters it.

Example

A service assistant may only retrieve from released manuals and service bulletins in the current versions; drafts, forum posts and earlier generated answers are excluded.

How it applies

  • Allowed sources: The corpus of an answer function names its sources: which document types, which status (released only), which versions and which languages. This is a data governance decision with an owner.
  • Exclusions: Drafts, AI-derived content that has not been released, and foreign text (customer uploads, web pages, supplier documents not taken over into the company's own release process) are kept out or kept in separate, labeled collections.
  • Maintenance: New releases are added, changed content is replaced and retired content is removed, so that the corpus follows the content lifecycle.
  • Security: A corpus with controlled entry points is harder to poison. OWASP lists manipulated retrieval data and weaknesses of vector stores among the main risks of LLM applications.
  • Limits: Curation makes better answers possible but does not guarantee them; a model can still misread or combine correct sources wrongly, so grounding and source citations remain necessary.

Curated corpus vs. source of truth

The source of truth is the authoritative place where content is maintained. A curated corpus is a selection or copy prepared for an AI function; it must be kept in step with the source of truth and never edited on its own.

External references