Semantic Chunking
Semantic chunking splits content into passages along meaning boundaries rather than fixed lengths, so each chunk is a coherent, self-contained idea that retrieves and cites cleanly.
Semantic chunking cuts content where ideas naturally end. Instead of slicing a page every N words, it detects topic boundaries and groups text into coherent units — so each chunk is one complete thought rather than an arbitrary fragment that might start or stop mid-idea. The result embeds more cleanly and retrieves more accurately.
You influence this directly through structure. Content already organized into self-contained, well-headed sections gives a semantic chunker clean lines to cut on, producing chunks that stand alone as answers — the extractability pillar doing double duty. A wall of text forces the system to guess at boundaries, and guesses produce weaker chunks. In effect, writing one clear idea per section is pre-chunking your own page the way you'd want it cut.
Example. A guide with distinct headed sections — "what it costs," "how long it takes," "what you need" — chunks neatly into three quotable passages. The same content as one undifferentiated block risks being split mid-sentence, weakening every chunk.
Relevant pillar
Related terms
- ChunkingChunking is how a retrieval system splits your page into smaller passages before indexing it, so AI engines retrieve and cite chunks of a page rather than the whole document.
- EmbeddingsEmbeddings are numerical representations of text that capture its meaning, letting AI systems find passages that are semantically related to a query even when they share no exact keywords.
- Passage RetrievalPassage retrieval is the practice of finding and returning specific relevant passages from within documents, rather than whole pages, which is why AI engines cite paragraphs instead of articles.