---
language: en
title: "Custom corpus"
description: "A custom corpus is a curated collection of searchable documents. Selection, updates, and permissions determine which evidence a response can retrieve."
canonical: "https://tropk.ai/en/resources/glossary/custom-corpus"
last_modified: "2026-10-12T16:13:15.234Z"
---

# Custom corpus

Reviewed on 2026-10-06.

## What it means

A custom corpus is a curated collection of documents that a system can consult to answer questions. It can bring together manuals, specifications, or internal materials selected for a particular use. Google’s patent application on the subject describes an external application that retrieves documents from this collection and provides data to condition a language model’s response.

## Why it matters for marketing

For marketing, the collection defines which evidence is available to an organization’s assistant. If old versions of presentations and policies coexist without identification, the response may use outdated commercial terms. Choosing sources, assigning owners, and reviewing permissions are all part of content work. Publishing material online does not automatically add it to this selected collection.

## Example in practice

Imagine a filter manufacturer that builds an assistant for resellers using parts catalogs. A test retrieves the 2022 manual and gives a connection that is incompatible with the 2026 model. The team finds that the new catalog was left out of the custom corpus. It decides to add it with a version label and restrict the old document to the corresponding products.

## How to apply or verify

Check the collection inventory, inclusion criteria, versions, permissions, and update schedule. Run questions with known answers and check which documents were retrieved, when that evidence is available. The team responsible for integration should test exclusions and access across profiles. Curation validates the sources; response evaluation checks how they were used.

## Important distinctions

A custom corpus refers to the set of sources, while [RAG](/en/resources/glossary/rag) describes an approach that combines [retrieval](/en/resources/glossary/information-retrieval) and generation. Consulting documents does not mean incorporating them into the model’s parameters or guarantee that every response will use the collection. The cited patent describes architectural possibilities; it does not prove that any public Google feature works this way or consults the same documents.

## Sources and references

- [Query response using a custom corpus — US20240362093A1](<https://patents.google.com/patent/US20240362093A1/en>) — Google LLC, via Google Patents

## Continue a leitura

- [Non-parametric memory](</en/resources/glossary/non-parametric-memory>)
- [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>)