Chunking strategies

How you split documents decides what can be retrieved. Chunks must be small enough to be precise and large enough to be self-contained.

Evolving: the basics hold, details shift. Reviewed 17 Oct 2026.

Key ideas

• Start with structure-aware chunking: split on headings, sections, list items, table rows.
• Typical sizes 300 to 800 tokens with 10 to 20 percent overlap; tune on your eval set.
• Parent-child: retrieve small chunks, return the larger parent section to the model.
• Contextual chunking: prepend a short LLM-written summary of where the chunk sits in the document before embedding.
• Keep tables intact and convert them to markdown; store page numbers and section paths as metadata.

Common mistakes

• Fixed character splits that cut sentences, tables and code blocks in half.

Go deeper

• Contextual Retrieval (Anthropic)