Skip to main content
Technical implementation · AI Search Infrastructure

​
Definition

Overlap is a technique in fixed-size chunking where adjacent chunks share a set number of tokens at their boundaries. A chunk beginning at position 500 and ending at 1000 might share tokens 950–1000 with the next chunk beginning at 950. Reduces the chance of a relevant sentence falling at a boundary and being lost across two incomplete chunks. The tradeoff: larger indexes and the possibility of retrieving the same content twice from adjacent overlapping chunks. Standard overlap windows run 10–20% of chunk size. For content strategy, overlap is a mitigation for poor structure — self-contained sections reduce its necessity.

Fixed-size chunking

Chunking

Embedding

RAG

Vector database

​
Relevant Plate Lunch Collective Services

AI SEO