Passage ranking

Passage ranking is the shift from treating a page as the atomic unit of retrieval to treating its individual passages as candidates in their own right. Google introduced passage-based indexing into classical search around 2020-2021; the same logic is intrinsic to how AI search works, because the chunking and vector retrieval underneath RAG operate on passages, not whole documents.

The consequence is that a single excellent passage buried in an otherwise unremarkable page can be retrieved and surfaced, and conversely, a strong page can fail to surface for a specific question if no single passage answers it cleanly. The unit of competition is the passage.

For optimization this reinforces the self-contained sentence discipline: write passages that answer a specific question completely on their own, put the direct answer where it does not depend on the surrounding page, and structure each section around a discrete claim. A page that is a collection of strong, independently-answerable passages will be retrieved across many more queries than a page that has to be read whole to make sense, which is also the mechanism behind query fan-out reaching deep into a site.