What is Chunk retrieval?

Chunk retrieval is the step in an AI search pipeline where a page is split into passages and individual passages, rather than whole documents, are scored and selected as candidates for an answer.

It is the mechanical reason passage-level structure matters more for AI visibility than page-level optimisation: the unit that competes is the chunk, so a strong page whose relevant passage is buried inside a long undifferentiated block competes worse than a weaker page whose passage stands alone. Chunking strategies vary by system, but most split on structural boundaries — headings, paragraphs, list items — and then embed each chunk for semantic matching against the query or sub-query. Two consequences follow for anyone writing content. A passage needs to make sense without the surrounding context, because it will be evaluated and possibly quoted in isolation, which means resolving pronouns and repeating the subject rather than relying on an earlier sentence. And heading structure does real retrieval work rather than merely organising the page for readers, since headings are where the boundaries usually fall.