How AI answers are made
Chunking
Chunking means cutting long documents into smaller pieces of text before they are indexed, so a retrieval system can return just the passage that answers a question.
In plain words
A whole web page is often too long to hand to a model in one go, so RAG systems split it into chunks first.
Anthropic describes chunks as usually no more than a few hundred tokens each.
Example
A long guide to bike repair could be split into chunks such as "fixing a flat tyre", "adjusting the brakes" and "oiling the chain". A question about brakes then retrieves only the brakes chunk.
Why it matters for AI search
Chunking is about how AI systems process your page, not something you have to do yourself.
Google says you don’t need to break your content into tiny pieces for Google Search. Its systems understand several topics on one page and can show the relevant part.
What you can do
- Write for people first: one clear heading per subtopic already creates natural chunks.
- Make the first sentence under each heading answer that heading.
- Don’t split a good page into many thin pages for AI. Google warns that making separate pages for every query variation, mainly to manipulate rankings, violates its scaled content abuse policy.
Often confused with
- Passage ranking
- Chunking is how a retrieval system splits documents. Passage ranking is a Google system that judges individual sections of a page.
Sources
- Anthropic: Introducing Contextual Retrieval
- Google Search Central: Optimizing your website for generative AI features on Google Search
Facts and sources last checked on 19 October 2026.
Related terms
How does your page do in AI search?
GEO-Score checks one page on 22 signals, from AI crawler access to structured data, and shows what to fix first.
Analyze a page