AI engines extract 40-120 word self-contained chunks from pages to use as direct answers in AI-generated responses. The specific chunk structure that makes a page reliably extractable has four properties: a question-format heading signaling what the chunk answers, a direct-answer first sentence stating the specific answer immediately, specific elaboration in 2-4 following sentences expanding the answer with named entities and specific numbers, and self-containment such that the chunk reads correctly without surrounding context. Pages built with specifically this chunk structure get cited at 3-5x the rate of pages with the same content presented as connected narrative, and the specific structural shift is the primary AEO investment most brands can make.
AI engines extract 40-120 word chunks. Four properties: question-format heading, direct-answer first sentence, specific elaboration with entities and numbers, self-containment. Pages with chunk structure cited 3-5x more than narrative-structured pages.
What specifically is a chunk and how does an AI engine use it?
A chunk is a self-contained section of a page that an AI engine can lift as a direct answer to a specific user query without needing to rewrite or synthesize across multiple sections. The AI engine's citation workflow processes queries by scanning crawled pages for extractable chunks that answer the specific query, selecting the best-matched chunk from the best-authority source, and presenting the specific chunk as the direct answer with attribution to the source page. Pages organized as a sequence of chunks produce more citations than pages organized as flowing narrative because the specific chunks map directly to specific user queries, which is the specific match the AI engine is looking for.
Why is the 40-120 word range the specific sweet spot?
The 40-120 word range is the specific sweet spot because it balances specific completeness against specific extractability. Chunks under 40 words typically lack enough specific elaboration to serve as a complete answer — they state the answer but do not provide the specific supporting detail that gives the answer credibility. Chunks over 120 words typically contain multiple distinct ideas that AI engines cannot cleanly extract as a single answer — the specific extraction either lifts only part of the chunk (losing specific context) or lifts the whole chunk (which is too long for the AI response format). The 40-120 range gives AI engines a complete answer that fits the specific response format, which is specifically why chunks in this range get cited more reliably than longer or shorter alternatives.
What specifically counts as a question-format heading for chunk signaling?
A question-format heading is a heading phrased as a specific question that a specific user might type into Google, ChatGPT, or Perplexity — ending with a question mark and matching the specific semantic shape of a search query. 'What is a good DTC conversion rate in 2026?' is a question-format heading. 'DTC Conversion Rates' is a topic label, not a question. The specific question-format signals to the AI engine what specific query the chunk answers, which the engine uses to match the chunk to specific incoming queries. Pages with topic-label headings require the AI engine to infer what each section answers, which is less reliable than explicit question-format signaling. The specific heading format matters because it is the specific first signal the engine processes when deciding whether to extract the chunk.
What does the direct-answer first sentence specifically look like in practice?
The direct-answer first sentence states the specific answer to the specific heading question in the first sentence, before any elaboration or context. For a heading 'What is a good DTC conversion rate in 2026?', the direct-answer first sentence is 'A good DTC conversion rate in 2026 sits between 2.5 and 4.5 percent for established brands.' The specific answer appears immediately — not after context, not after a transition sentence, not after a general topic introduction. AI engines preferentially extract the first sentence after a question heading because that is where direct answers reliably live in well-structured content. The specific pattern makes extraction almost automatic for the AI engine, which raises the specific citation probability. AI-referred visitors arrive pre-sold specifically because the AI engines were able to extract clear answers from the specific source pages.
What makes a chunk self-contained enough for extraction?
A chunk is self-contained when it reads correctly without the surrounding paragraphs. The specific test: paste just the chunk into a document and read it. If the chunk makes complete sense — the specific answer is clear, the specific entities and terms are named rather than referenced ('this product' or 'the above metric' pointing to something outside the chunk), the specific numbers and claims stand alone — the chunk is self-contained. If the chunk requires context from earlier in the page to make sense ('as mentioned above', 'this framework we covered'), it is not self-contained and will not extract cleanly. The specific self-containment discipline requires writing each chunk as if it might be the only part of the page a reader ever sees, which is specifically what AI engines will often make true.
How do you audit your pages for chunk structure?
1) Open the page and identify each H2 or H3 section. 2) Check the heading: is it a specific question that a user would type into a search engine? If no, rewrite as a question-format heading. 3) Read the first sentence after the heading: does it directly answer the heading question? If no, rewrite the first sentence as a direct answer. 4) Read the full section: is it 40-120 words of specific elaboration with named entities and specific numbers? If less than 40, add specific supporting content; if more than 120, split into two sections. 5) Read the section in isolation: does it make complete sense without surrounding context? If no, rewrite to remove dependencies on earlier sections. 6) Deploy the restructured page and monitor AI engine citations over the following 8-12 weeks. The specific improvement typically shows as more frequent citations across more specific queries, which is the specific AEO lift chunk structure produces.
Conclusion
Chunk structure is the specific technical discipline that makes a page extractable by AI engines, and most pages built for Google ranking do not naturally produce it. The practical next step for any team currently publishing content without specific chunk discipline is to audit the top 10 highest-traffic pages against the four properties and queue the specific restructuring work. In Advize's experience, specific chunk-structure improvements typically produce AI citation lift within 8-12 weeks as the AI engines re-crawl and re-index the restructured content.