What Is Chunking in RAG?

Chunking splits large documents into smaller pieces so a RAG system can retrieve focused context.

Chunking is the process of breaking documents into smaller sections before indexing them. Each chunk can be embedded, searched, and passed to the model as context.

In practice

Good chunks are large enough to keep meaning but small enough to retrieve precisely. Headings, paragraphs, tables, and document structure can all affect chunk quality.

What to watch

Bad chunking causes bad answers. Tiny chunks lose context, while huge chunks can bury the relevant sentence inside too much text.