Passage chunking · Text splitting
Chunking is the splitting of a page into semantically coherent passages before vector indexing. Each chunk is then scored independently: a page can be ignored as a whole yet cited for a single one of its paragraphs.
Standalone definition · 37 words
How does chunking work?
An AI search system splits text into blocks, by headings, paragraphs or a fixed number of tokens, then computes an embedding for each block. At query time, it compares the question’s embedding with those of the blocks and retrieves the closest ones. Chunking strategies vary by system, and no mainstream engine publishes its own.
Why does chunking change how you write?
A chunk must make sense without the rest of the page. A paragraph that relies on “as explained above” loses its meaning once isolated. Each section should name its subject, contain its answer and stay short.
How do you optimize a page for chunking?
Question-style headings, the answer at the top of the section, one topic per section, tables with explicit headers and short lists. Avoid blocks of several hundred words with no subheading, which get split with no editorial logic.
Frequently asked questions
What is the ideal chunk size?
There is no documented size for mainstream engines. In in-house RAG systems, common sizes range from a few hundred to a thousand tokens, depending on the use.
Does Google split pages into passages?
For ranking, yes: Google announced passage ranking in 2020, which evaluates specific passages of a page. This says nothing about the method used by AI Overviews. Its May 15, 2026 guide is explicit: no splitting into small pieces is required for Google Search, whose systems understand several topics on one page. The value of self-contained passages is argued and measured for other systems, notably by iPullRank and in Metehan Yeşilyurt’s observations of ChatGPT.
Does chunking matter for classic SEO?
Indirectly. The same writing practices serve passage ranking, featured snippets and LLM retrieval.
Related terms
Further reading
- Is Chunking a technique to optimize SEO and GEO ?
- Multimodal Geo Audit with Gemini-Embedding-2
- Microsoft Clarity, Bing Webmaster Tools: the Only GEO Stack That Actually Measures LLM Visibility
- ← Back to the GEO glossary (114 terms)