AI answer engines do not read your page the way a human does — they split it into small passages called chunks, convert each chunk into a numeric vector, and store those vectors so they can later fetch the few chunks that best match a user's question. That means the unit that gets retrieved and cited is not your page; it is a passage. If your best answer is buried in the middle of a long, rambling section, the system may never surface it. Writing for retrieval is the practice of making each passage a clean, self-contained, findable answer.
What chunking and embeddings actually do
When a retrieval-augmented system ingests your page, it breaks the text into chunks — often a few hundred words, frequently along heading or paragraph boundaries. Each chunk is passed through an embedding model that maps its meaning to a vector, a list of numbers positioned so that semantically similar text sits nearby. At query time, the user's question becomes a vector too, and the system returns the chunks whose vectors are closest. The model then writes its answer from those retrieved chunks and cites the pages they came from.
The practical consequence: relevance is judged passage by passage, not page by page. A page can rank well in classic search yet lose in AI retrieval because its winning idea is diluted across a wall of text. Your goal is to make sure the chunk that contains your answer is coherent on its own and unmistakably about one thing.
Structure your page into clean, retrievable passages
Because chunk boundaries tend to follow structure, structure is your main lever. Give every distinct idea its own section with a descriptive H2, and open that section with a one- or two-sentence direct answer before you elaborate. Keep paragraphs short so a single passage carries a complete thought without dragging in unrelated material. Use lists for steps and criteria, because a tight list is easy to extract cleanly.
- One idea per section, introduced by a question-style H2.
- Answer first, context second — front-load the sentence a model would quote.
- Short paragraphs so a chunk does not blend two topics.
- Define terms in place; a chunk retrieved alone should still make sense.
This is the same answer-first discipline that wins featured snippets, so it pays double. Our guide to FAQ and How-To Schema shows how to reinforce these passages with structured data that makes the question-and-answer shape explicit to machines.
Write self-contained chunks, not paragraphs that depend on each other
The biggest retrieval killer is context that lives outside the passage. If a paragraph says "as mentioned above" or relies on a pronoun whose antecedent was three paragraphs back, that chunk becomes ambiguous once it is pulled out on its own. Repeat the key noun instead of leaning on "it" or "this." Restate the subject at the start of a new section. Avoid answers that only make sense after reading the whole page. A good test: copy any single paragraph, read it cold, and ask whether it still answers a real question. If not, it will not retrieve well.
Keep the underlying HTML clean too. Retrieval only works on text the system can actually extract, so if your content renders only after client-side JavaScript, the chunker may get nothing. We explain that failure in depth in Why JavaScript-Only Websites Are Invisible to AI Assistants.
Help the system find the right chunks
Two habits raise the odds that the correct passage is retrieved. First, mirror the language your audience uses in headings and opening sentences, because embeddings match meaning and phrasing; a heading worded like a real question sits close to that question's vector. Second, build clear internal links between related passages so the system understands your content as a connected topic rather than isolated fragments — the approach in Internal Linking for SEO and AI applies directly at the chunk level.
You do not need to know the exact chunk size any given engine uses. Optimizing structure, self-containment, and phrasing works across all of them because they all reward passages that are coherent and on-topic.
Turn your page into a set of answers
Reread your most important page and mark where each answer lives. If a key answer is stranded mid-paragraph, promote it: give it a heading, lead with the direct statement, and make the surrounding sentences stand on their own. You are not dumbing the page down — you are packaging your expertise so the machine can lift the right piece and credit you. To confirm an AI crawler can actually read and extract those passages from your live site, run the CheckMy.site scanner and fix anything blocking clean extraction.