Passage ranking, or chunk retrieval, is when a search or AI system scores individual sections of a page rather than the whole page, so one strong passage can be surfaced or cited on its own.
What is Passage Ranking (Chunk Retrieval)?
Passage ranking is a search technique that evaluates individual passages within a page, not only the page as a whole. In AI search the same idea is called chunk retrieval: documents are split into chunks, each chunk is scored against the query, and the best chunks are sent to the language model. A single clear section can win even on a broad page.
How it works
Google announced passage ranking in October 2020 and has described it as a way to find the needle-in-a-haystack answer buried in a long page. The page is still indexed as a whole, but ranking can lean on the most relevant passage.
AI search engines take this further. In a typical RAG pipeline:
flowchart LR
A[Page] --> B[Split into chunks by heading or length]
B --> C[Score each chunk against the query]
C --> D[Top chunks go to the LLM]
D --> E[Answer cites the source page]
Chunks are often split at headings or by length. A chunk that starts with “As mentioned above, it also…” loses its meaning when separated, while one that begins with a clear statement survives on its own.
Why it matters for AI search
Engines cite pages based on the passages that support a specific sentence in their answer. Content structured as self-contained, answer-first sections gives engines more usable chunks and more chances to be cited.
Examples
- A long guide to cloud security posture management has a section titled “How CSPM differs from CNAPP”. That section alone is retrieved and cited for a comparison prompt.
- A FAQ block with one question per heading produces clean chunks that map neatly to user questions.
- A product page with key facts only in images gives the retriever nothing to score.
Chunking also explains why very long pages are not automatically better. Depth helps only if each section is focused and clearly labeled; a long page with vague headings produces chunks that score poorly against any specific question.
How to apply it
- Use descriptive H2s and H3s that state the question or topic.
- Open each section with a direct, one- or two-sentence answer.
- Repeat the subject name instead of relying on “it” or “this” from earlier sections.
- Keep facts in text, tables and lists rather than images.
- Cover one idea per section.
Related terms
Frequently asked questions
Does Google index passages separately?
No. Google has said it still indexes whole pages, but its ranking systems can consider the relevance of specific passages within a page.
What is a chunk in AI search?
A chunk is a section of a document, often a few paragraphs, that a retrieval system splits out and scores on its own before passing the best ones to a language model.
How long should a passage be?
There is no fixed rule. Aim for sections that answer one question completely, start with the answer, and make sense without the rest of the page.
