# What is Chunking?

Source: https://mushbloom.uk/resources/glossary/what-is-chunking.html

By Mushbloom. Published and updated: 2026-10-12.

## Definition & overview

Chunking splits documents into smaller passages for retrieval or model processing. Useful boundaries preserve meaning and provenance; arbitrary cuts can separate an answer from its conditions or source.

## Analogy & mental model

You file a handbook by complete sections rather than tearing each page into equal strips.

Keep headings, source identifiers and necessary surrounding context with each passage.

## Why it matters

Retrieval quality depends partly on what a passage contains. A chunk that omits a heading or qualification can look relevant while supporting the wrong answer. Good boundaries keep meaning, provenance and enough context together without overwhelming the model.

## Practical business example

A policy exception stays with its eligibility conditions, rather than becoming a standalone passage suggesting the exception applies to every customer.

Illustrative scenario, not a client case study or promised result.

## How it works

1. Split source material along meaningful sections or logical units.

2. Preserve headings, identifiers and needed surrounding context.

3. Test questions whose answers depend on conditions crossing a potential boundary.

## Core components

- Passage boundaries

- Source metadata

- Context preservation

Equal-size splitting is not automatically suitable for every document, table or policy.

## Chunking vs Context window

Chunking prepares candidate passages for retrieval. The context window limits how much selected material the model can consider in one interaction.

## Frequently asked questions

What does Chunking mean?
Chunking splits documents into smaller passages for retrieval or model processing. Useful boundaries preserve meaning and provenance; arbitrary cuts can separate an answer from its conditions or source.

How is it different from Context window?
Chunking prepares candidate passages for retrieval. The context window limits how much selected material the model can consider in one interaction.

What is a sensible first step for a small business?
Keep headings, source identifiers and necessary surrounding context with each passage. Split source material along meaningful sections or logical units.

What limitations should I keep in mind?
Equal-size splitting is not automatically suitable for every document, table or policy.

## Sources

Google Cloud: Retrieval-augmented generation: https://cloud.google.com/use-cases/retrieval-augmented-generation

No rankings, AI citations or commercial outcomes are guaranteed.