Decoder. plain-English AI glossary

Passage Retrieval

● Core

Also called Passage Search

Fetching the specific paragraphs that answer a question, not whole documents.

Think of it like

Highlighting the exact sentences in a report instead of handing someone the entire binder.

Example

Asked about a refund window, the system returns the one paragraph stating "30 days", not the full 40-page policy.

How it actually works

Whole documents are too big and too diluted to hand a model; passage retrieval works at the paragraph or chunk level so the relevant bit lands in context. Granularity is the whole game: too coarse and you waste context on filler, too fine and you sever the sentence that gives an answer its meaning. It is the retrieval half of most question-answering systems.

For product teams

Precise passages mean tighter, better-grounded answers and lower token cost than dumping full docs.

For engineers

Chunk-level dense/sparse retrieval; answer quality hinges on chunk granularity and boundary choices.

Chunking — Depends on how documents are chunked.
RAG — The retrieval step feeding a RAG answer.

Read anything AI without the jargon

Look up any term in plain English, or save terms as you read with the free Chrome extension.

Open Decoder Add to Chrome