Skip to content
AEO Canon · the reference for answer-engine optimization

What Is Chunking? How AI Splits Your Page Before Reading It

Before an AI reads your page, a retrieval system splits it into chunks — passages divided by heading or token window — then embeds and retrieves the best match. A self-contained, answer-first passage under a clear heading survives chunking and gets cited; a buried answer split across chunks does not.

BBurke Atkerson2 min read

Before any AI "reads" your page, a retrieval system chops it into chunks — and it retrieves single chunks, not whole pages. That means a self-contained answer under a clear heading gets cited, while a brilliant answer split across chunk boundaries never surfaces.

Quick answer

Retrieval systems split your page into chunks — by heading/section or by fixed token windows — then embed each chunk and retrieve the ones that best match a query. So an answer-first, self-contained passage under a clear heading survives chunking and gets cited. A buried answer split across chunks does not.

What is chunking?

Chunking is the splitting step. A retrieval system breaks your page into smaller passages — often along heading and section boundaries, or by fixed token windows — so each piece is small enough to embed and compare. Your page never enters the system as one blob; it enters as a set of chunks.

Why does the chunk boundary decide citation?

Because retrieval happens at the chunk level. The engine embeds each chunk, matches the query against every chunk, and pulls the closest ones. If one chunk holds your full answer, it can be retrieved and quoted. If the answer is spread across two chunks — or buried mid-paragraph — no single chunk matches strongly, so none gets cited.

How do I write chunk-proof passages?

Make each passage stand alone. Put a complete answer directly under a clear, descriptive heading, near the top of the section, without relying on earlier paragraphs for context. If a passage still makes sense when lifted out of the page, it will survive chunking.

01Your pagefull article as published
02Split into chunksby heading or token window
03Embed each chunkone vector per passage
04Retrieve best chunkclosest match to the query
05Citedthe surviving passage is quoted
How a page becomes a citation. The engine matches and quotes individual chunks — so the answer must live complete inside one.

Write for the chunk, not just the page. More in the content format AI cites most, chunking, and semantic chunking.

What content format does AI cite most?

Self-contained, answer-first passages under clear question-shaped headings.

Read the full answer →
What is chunking?

The step that splits a page into passages before embedding and retrieval.

Read the full answer →
What is semantic chunking?

Splitting by meaning and section boundaries so each chunk is a coherent passage.

Read the full answer →

Frequently asked questions

What is chunking in AI retrieval?
Chunking is the step where a retrieval system splits your page into smaller passages before storing it. Systems chunk by heading or section, or by fixed token windows, then embed each chunk separately. When someone asks a question, the engine retrieves the best-matching chunks rather than the whole page.
Why do chunk boundaries decide whether I get cited?
Because the engine retrieves and cites individual chunks, not whole pages. If your answer sits complete inside one chunk under a clear heading, that chunk can be retrieved and quoted. If the answer is split across two chunks or buried in a long block, no single chunk contains the full answer, so none matches well enough to be cited.
How do I write content that survives chunking?
Put a complete, self-contained answer directly under a clear, descriptive heading, and keep it near the top of that section. Avoid answers that depend on earlier paragraphs for context, because chunking may separate them. Each passage should make sense on its own if lifted out of the page.

Related reading

AI-detection tools are unreliable and not what answer engines use to decide citations — engines judge content on quality, originality, and accuracy, not on whether a machine wrote it. Stop chasing detection and make content genuinely original, because generic content fails no matter who wrote it.

2 min read

AI-generated content can get cited, but only when it's made genuinely original, accurate, and useful — raw model output tends to be generic, unsourced, and interchangeable, which is exactly what engines skip. The deciding factor is the substance and originality you add, not whether a model helped write it.

2 min read