---
language: en
title: "Passage retrieval"
description: "Passage retrieval selects relevant excerpts for a question. Learn how to evaluate the context retrieved before judging an AI response."
canonical: "https://tropk.ai/en/resources/glossary/passage-retrieval"
last_modified: "2026-10-12T16:13:14.434Z"
---

# Passage retrieval

Reviewed on 2026-10-06.

## What it means

Passage [retrieval](/en/resources/glossary/information-retrieval) is the selection of document excerpts that may help answer a question. The retrieved unit is a passage, rather than necessarily an entire page. The mechanism may compare words or vector representations. Its role is to provide candidate context; producing the final answer is usually a later step in the system.

## Why it matters for marketing

For marketing, this distinction helps assess a knowledge base without attributing every failure to the generated text. A page may contain the correct information, but the system may select a different passage. The team needs to check which excerpt made it into the answer and whether it preserves the commercial terms, exceptions, and references needed to interpret the offer correctly.

## Example in practice

Imagine a hotel chain whose policy combines cancellation, payment, and check-in times. When asked about canceling non-refundable bookings, the assistant retrieves only the paragraph about flexible rates. A review finds the correct condition in another passage in the same document. The team decides to investigate passage selection before changing the public policy.

## How to apply or verify

Create representative questions and identify in advance the passages that contain sufficient evidence for each answer. Ask the search owner to provide a record of the returned passages, their positions, and their source documents. Compare the selection with this reference set. Separately assess whether the generation used the context, since finding the right passage alone does not prove the answer is correct.

## Important distinctions

Passage retrieval is not synonymous with public [citation](/en/resources/glossary/ai-citation) or page ranking in Google Search. [Chunking](/en/resources/glossary/chunking) prepares units that can be retrieved; retrieval selects among those units. [RAG](/en/resources/glossary/rag) uses retrieved information to support generation. None of these steps guarantees that a passage is up to date or that its claims are true.

## Sources and references

- [Dense Passage Retrieval for Open-Domain Question Answering](<https://aclanthology.org/2020.emnlp-main.550/>) — Association for Computational Linguistics
- [Chunk large documents for RAG and vector search in Azure AI Search](<https://learn.microsoft.com/en-us/azure/search/vector-search-how-to-chunk-documents>) — Microsoft

## Continue a leitura

- [Chunking](</en/resources/glossary/chunking>)
- [RAG](</en/resources/glossary/rag>)
- [Information retrieval](</en/resources/glossary/information-retrieval>)

## About the author

[Vinicius Pacheco](</en/team/vinicius-pacheco>)

Responsible for Tropk’s technical content, focusing on crawling, indexing, structured data, and agent access to content.

---

[All pages in Markdown](<https://tropk.ai/sitemap.md>)