AI series · Part 2: Retrieval (RAG)
Chunking Strategies: How to Cut Documents Without Losing Meaning
Chunking splits big documents into searchable pieces. Cut in the wrong place and the answer loses half its meaning. Here are the five common methods and when to use each.
Chunking means cutting big documents into small pieces so each piece can be searched on its own. It's like handing someone the one page they need instead of the whole 300-page book. But if you cut in the middle of a sentence, the meaning is lost.
Five ways to chunk
| Method | What it means |
|---|---|
| Fixed size | Every N words, with some overlap |
| By structure | One chunk per heading or section (best for policies and manuals) |
| Semantic | Cut where the topic changes |
| Parent–child | Search small pieces, return the full section |
| Add a header | Put the title in every chunk, e.g. "[Leave Policy > Sick Leave]" |
Real-world patterns
- Law firms make one chunk per contract clause.
- Product manuals use one chunk per section, such as "How to reset your router".
Why structure wins for policies
Our HR assistant makes one chunk per policy section. With fixed-size cutting, "12 days sick leave" and "only after probation" could land in different chunks. The retriever would return the first, the answer would miss the probation rule, and an employee on probation would get the wrong answer.
I chunk by document structure and add the title and section to each chunk, so every chunk is precise and makes sense on its own.
- The short version
Next in the GenAI series: Embeddings: Google Maps Coordinates for Meaning
Start from the top: GenAI Quick Answers and Series Guide
We deploy these patterns across production engagements in fintech, healthcare, and eCommerce. Talk to a senior engineer about how they apply to your platform.