Chunking

Chunking is the process of splitting documents into smaller fragments before vectorization in RAG systems. This critical step directly impacts retrieval quality: chunks that are too long dilute semantic meaning, while chunks that are too short lose essential context. The optimal strategy splits content at H2 headings, creating autonomous chunks where each fragment stands alone. Chatbots typically use 200-300 word chunks, while content analysis systems use 400-600 words.

For SEO applications, chunking determines whether AI Search will cite your content—Google extracts fragments organized under H2 headings and evaluates them independently. This makes the self-contained chunk principle key: every H2 section needs BLUF at the start, full entity names rather than pronouns, and evenly distributed key terms throughout each chunk.