AI models and search systems have limits on how much text they can process at once. Chunking handles large documents by splitting them into sections — paragraphs, fixed-size blocks, or semantically coherent units — indexing each piece separately, and retrieving only the relevant chunks when a query comes in.
A 40-page supplier agreement split into 80 chunks of roughly 500 words each means that a question about the termination clause retrieves only the two or three relevant chunks rather than the entire document. But chunk size is a real trade-off: chunks that are too small lose context, and chunks that are too large reduce retrieval precision. Getting chunking right is one of the less-glamorous but genuinely important decisions in building a working RAG system.
