Content
How to write content an AI assistant will actually quote
16 July 2026 8 min read
Assistants quote passages, not pages. That single fact should change how you write. A model extracting an answer wants a chunk of text that stands alone, states something concrete and can be attributed without ambiguity. Most marketing copy fails all three tests because it builds to a point instead of leading with one.
Lead with the answer
Put the direct answer in the first 40 words under the heading, then explain. If your page is 'How much does X cost', the first sentence should contain a number or an honest range. Background, positioning and story go underneath.
Write self-contained paragraphs
Retrieval systems split pages into chunks. A paragraph starting 'This means you can…' is worthless out of context. Repeat the subject: 'Weekly re-grading means you can…'. It reads slightly heavier to a human and dramatically better to a machine.
Be specific enough to be checkable
Models favour text that carries verifiable detail: prices, counts, dates, named integrations, named jurisdictions. 'Fast setup' is unquotable. 'Setup takes about ten minutes and needs one DNS record' is quotable.
Structure that helps extraction
Use the shape of the page to signal where answers live.
- One H1 stating the page's subject as a noun phrase.
- H2s phrased as the questions buyers actually ask.
- Short answer paragraph directly under each H2.
- Tables for comparisons — they survive extraction better than prose.
- A summary block near the top for long guides.
- FAQ schema mirroring your on-page questions word for word.
The things that quietly hurt you
Undated posts, superlatives with no evidence, duplicate boilerplate on every page, and 'contact us for pricing'. Each one gives a model a reason to prefer a competitor's page that simply said the thing.
See where your own site stands
Run the free grader for a 100-point score across performance, SEO, mobile and security, plus an AI readiness rating and a prioritised fix list.
Grade my site