AI Glossary · Search & RAG
Chunking
Splitting source documents into passages sized for retrieval and the model's context window. Chunk size and overlap are top-3 RAG quality levers.
Definition
What is Chunking?
Chunking is splitting source documents into passages sized for retrieval and the model's context window. Chunk size and overlap are top-3 RAG quality levers.
- Category
- Search & RAG
- Glossary set
- 8 related terms
- Audience
- Enterprise AI leaders
Why does Chunking matter in enterprise AI?
Chunking matters in enterprise RAG because it affects retrieval quality, answer trust, citation accuracy, and how safely AI systems use governed knowledge.
Related terms in Search & RAG
- Enterprise Search
- Unified search across internal systems (SharePoint, wikis, tickets, CRM) — increasingly powered by embeddings and LLM answers with citations.
- GraphRAG
- RAG variant that retrieves over a knowledge graph and communities of entities; strong for multi-hop questions on connected enterprise data.
- Hybrid Search
- Combining keyword (BM25) and vector search to improve RAG relevance, especially for enterprise content with codes, IDs and jargon.
- RAG (Retrieval-Augmented Generation)
- Fetching relevant documents at query time and passing them to the model so answers cite trusted sources instead of relying on parametric memory.
- Re-ranker
- A second-stage model (cross-encoder) that reorders top-k candidates from initial retrieval for sharper relevance — the fastest RAG quality win.
- Semantic Search
- Search based on meaning rather than keywords, powered by embeddings; the retrieval half of most RAG systems.