mediumConceptDesignMidSeniorStaff
How do you choose a chunking strategy for a document corpus?
4 min readUpdated 2026-08-15
Answer
Short answer
Chunk boundaries decide what can ever be retrieved, so you derive them from the questions the corpus must answer and from the document's own structure, then confirm the choice with retrieval recall on a gold set rather than by eye.
Was this answer helpful?
Likely follow-ups
- Your corpus mixes two-page policy notes and 400-page manuals. Do they get the same strategy?
- How would you retrieve a small chunk but hand the model a larger window around it?
- What would convince you to re-chunk a corpus you have already indexed?
- Where does chunking stop helping and the question need structured data instead?
Related questions
- Your RAG system cannot answer a question whose answer is definitely in the documents. Where is the fault?hardAlso on rag and chunking4 min
- Your corpus is scanned PDFs, spreadsheets and slide decks. How do you ingest it?hardAlso on ingestion and document-parsing6 min
- How do you build a gold set for a retrieval system?mediumAlso on retrieval and rag5 min
- What metadata do you store alongside each chunk, and what is each field for?mediumAlso on ingestion and rag4 min