Skip to main content

Document performance

Analytics → Document Performance shows which documents are actually answering questions — and which are dead weight or in poor health.

Document performance

Per-document statistics​

For each document in your library:

  • Retrieval frequency — how often it appeared in search/answer results.
  • Average relevance — how strongly its passages matched when retrieved.
  • Top matching queries — what people were asking when this document surfaced.
  • Chunk health — how many extracted chunks are healthy versus suspect. A document with many suspect chunks may have extracted poorly (scanned pages, unusual layout) and is a candidate for reprocessing.
  • Reprocessing eligibility — failed documents can be re-run from here or from their details page.

How to use it​

  • A document is never retrieved — either nobody asks about its topic (fine) or its content isn't being matched (try clearer source material, or check how it was chunked on its details page).
  • Retrieved often but with low relevance — the document is adjacent to what people want but doesn't quite contain it; consider adding a more direct document.
  • Suspect chunks — open the document's details, inspect the chunks, and reprocess if extraction looks wrong. Adjusting extraction settings can help for whole classes of documents.

Next: Gap analysis →