← Concept Index

Retrieval & knowledge

Chunking & retrieval quality

Also called: splitting documents, why RAG missed it, retrieval failures

DEFINITION

Before documents can be retrieved they're split into chunks and indexed. How you split them — size, overlap, whether related text stays together — largely decides whether the right passage can be found at all.

WHY IT MATTERS

Most RAG failures aren't the model's fault; they're retrieval's. A fact split across two chunks, or buried in an over-large one, simply never reaches the model, which then answers confidently from memory instead. Fixing retrieval usually beats prompting harder.

COMMONLY CONFUSED WITH

Model quality. When a document-grounded answer is wrong, the instinct is to blame the model, but the passage often was never retrieved — a chunking-and-search problem, not a reasoning one.