What is Passage Ranking?
- Passage Ranking
- Passage ranking is the retrieval and scoring of individual sections of a page rather than the document as a whole, so one paragraph can be judged relevant to a question even when the article around it is about something broader. Passage ranking is why a long page can win a narrow question, and why the unit worth editing is the paragraph rather than the article.
01 In practice
Search engines and answer engines both work at passage level, for the same reason: a question usually has its answer inside a document rather than spread across all of it. Google announced passage-based ranking for search in 2020, and retrieval for language models operates on chunks of text by construction.
Writing for passage ranking means letting each section carry its own context. A paragraph that opens by naming its subject can be scored, retrieved and quoted on its own; a paragraph opening with a pronoun pointing at the previous section arrives at the model as a statement about nothing.
Headings do less work here than writers expect. A heading is often stripped from the text beneath it during extraction, so a section whose subject exists only in its heading loses that subject at exactly the moment it needed to survive.
02 Related terms
The terms next to this one.
Content
Content Chunking
Structuring a page into sections that each stand alone.
Technical
Retrieval
A model fetching live pages while composing an answer.
Core
AI Citation
When an assistant attributes part of its answer to your page.
Technical
Vector Embedding
Text turned into numbers, so meaning can be compared by distance.
The work behind it
Rank on Google. Get cited by ChatGPT.
The sources assistants quote and the pages search still ranks, one engagement, both surfaces.