techniques
Chunking
How engines split long pages into retrievable pieces — H2 boundaries are the dominant split signal.
Definition
Chunking is the process of splitting a long document into smaller pieces (200-1000 tokens each) so each piece can be independently embedded and retrieved. Most engines chunk at heading boundaries (H1/H2/H3), some chunk by paragraph, some use semantic-similarity-based splits. A page with strong H2 structure yields cleaner chunks; a wall-of-text page chunks awkwardly and underperforms.
Example
A 3000-word page with 6 H2 sections chunks into 6 retrievable pieces, each independently citable. A 3000-word page with no H2s chunks into ~6 arbitrary 500-token pieces that often split mid-thought.How to optimize
Use H2s every 400-600 words. Make each H2 section self-contained (don't reference earlier sections by 'as discussed above' — engines retrieve sections in isolation). VectorCite's h2-coverage signal scores this.