AI assistants quote passages, not whole pages. To be quoted, write each section so it answers one question on its own: put the answer in the first sentence, keep one idea per section, define terms plainly, name things instead of saying "it", and back facts with dates and sources. Content written this way is also easier for people to read.
This post explains why that works and gives a checklist you can apply to any page.
Why passages matter more than pages
Most assistants that answer from the web use retrieval. They split pages into chunks, usually along headings and paragraphs, compare each chunk with the question, and pass the best matches to a language model that writes the answer. The model then cites the sources it used.
So the competition is not your page against another page. It is your best paragraph against every other paragraph on the topic. A long, strong article can lose to a weaker page that has one clearer passage. We build these retrieval pipelines for clients, and the same rule applies inside a company as on the open web: see our retrieval and data service.
Put the answer first
Open each page, and each section, with the direct answer. Then add the detail, the caveats and the examples. Many writers do the reverse: a scene-setting intro, then background, then the answer near the end. For a reader in a hurry, and for a system matching a question, the answer is then buried in a chunk that is mostly about something else.
A useful test: if someone read only the first two sentences of the section, would they have the answer? If not, move it up.
One idea per passage
Give each idea its own heading and its own short block of text. A section that covers pricing, setup and support at once matches none of those questions well. Three short sections each match one of them clearly.
- Keep paragraphs short, usually two to four sentences.
- Phrase headings the way people ask: "How long does setup take?" beats "Timeline considerations".
- Use a list for steps and a table for comparisons. Both keep related facts together in one chunk.
Define terms and name things
A passage lifted out of your page loses its context. Words that point back to earlier text, such as "it", "this approach" or "the tool", may make no sense on their own.
- Name the entity. Write "llms.txt is a proposal for..." rather than "It is a proposal for...".
- Define terms on first use. One plain sentence: "Answer engine optimization (AEO) is the work of making content easy for AI assistants to quote."
- Be consistent. Use one name for a product or concept throughout. Switching between synonyms makes matching harder.
Structured data helps here too. JSON-LD states who you are and what the page is about in a form machines do not have to infer. See FAQ and structured data for answer engines.
| Hard to quote | Easy to quote | |
|---|---|---|
| Opening | A long scene-setting intro | The answer in sentence one |
| Section | Several ideas mixed together | One idea per section |
| Terms | Jargon with no definition | A plain definition on first use |
| References | It, this, the product | The named product or entity |
| Facts | Undated claims | Dated, sourced facts |
Facts, freshness and sources
Assistants that cite sources have a reason to prefer passages that look reliable. You cannot control their judgment, but you can remove reasons to doubt you:
- Date your facts. Show a last updated date on the page and put dates on time-sensitive claims.
- Cite primary sources. Link to the specification, the official documentation or the original data, not to a summary of a summary.
- Do not overclaim. Invented statistics and vague superlatives are exactly what careful systems and careful readers learn to ignore.
- Keep one version of the truth. If a fact appears on several pages, keep them in sync. Conflicting versions give the system a reason to cite someone else.
Make the page easy to read by machines
Good writing only counts if the text can be fetched. Check that:
- the main text is in the HTML, not injected later by JavaScript that some crawlers do not run;
- the crawlers you want are allowed in
robots.txt(see robots.txt for AI crawlers); - you offer a plain version where you can, such as Markdown on request and an llms.txt file.
A before and after
Here is a typical opening, followed by a rewrite.
Before: "In an ever-changing digital landscape, businesses face many choices when it comes to managing their online presence, and understanding the options available has never been more important."
After: "A staging site is a private copy of your website where you test changes before they go live. Use one for any change to checkout, payments or navigation."
The first version says nothing a reader or a system can use. The second defines the term, names the entity, and gives a direct recommendation in two sentences. If an assistant is asked "what is a staging site", the second version is a complete, quotable answer. The first is filler that dilutes whatever chunk it lands in.
Go through your most important pages and look for openings like the first one. Replacing them is usually the fastest content improvement you can make.
A passage checklist
Run through this for every important page:
None of this is a trick. It is the same advice that makes content useful to a person skimming on a phone. That is the point: answer engines are built to find content that helps people, so writing for people is the most durable way to be quoted. For the bigger picture, read what is answer engine optimization and AEO vs SEO vs GEO.
Next step
Content is half the job; the other half is making sure AI systems can reach and read it. Run the free agent-readiness scan to check your crawler rules, llms.txt, Markdown and structured data in one pass. For help with content operations, see our SEO service.