How to write a page an AI answer will quote

There is one pattern that does most of the work, and it takes ten minutes per page.

The pattern

That is it. Everything else on this page is detail.

Why it works

An extraction system is looking for a self-contained statement that answers a question without needing the rest of the page. If it finds one, it uses it and cites you. If it does not, it takes something from a page that has one, or writes its own summary and cites nobody.

Most pages make it fail. They open with context, then background, then a story, then the answer in paragraph four. A parser takes the first plausible thing it finds, and paragraph one is rarely it.

Writing the answer paragraph

Self-contained. No "as we saw above", no "this", no "it" pointing at an earlier noun. Somebody should be able to lift the paragraph out and have it still make sense.

Direct. The first sentence answers. Qualifications come after.

Specific. One real number, dated if it can be. "Fast" is not extractable. "Returns in under 40 milliseconds at the 95th percentile, measured in September 2026" is.

Complete but short. 40 to 60 words. If it needs a second paragraph, it is not one answer.

Bad:

Search relevance is a complex topic with many facets. Different teams approach it differently depending on their needs and constraints. In this article we will explore several approaches and discuss the trade-offs involved.

Good:

Typo tolerance means the search returns the right result when the query is misspelled. It needs an index that stores character sequences rather than whole words, usually a trigram index. Postgres supports it through the pg_trgm extension. Dedicated search engines enable it by default.

Writing the heading

Use the question. Not the topic.

Topic headings are a habit from print. Question headings match what gets typed.

The other extractable shapes

Tables. Models extract tables reliably. Any comparison, any set of options with attributes, any group of numbers should be a table rather than prose. A five-row table communicates more and survives extraction better than four paragraphs.

Numbered steps. For anything procedural. One action per step. No step that contains two actions.

Definition sentences. "X is Y that does Z." Boring and highly quotable.

Short lists with bold leads. Each item starting with the term in bold, then the explanation.

The markup

Add JSON-LD for what the page actually is: FAQPage if there is a visible FAQ, HowTo for a procedure, Article otherwise, plus BreadcrumbList and Organization.

One rule that matters more than the markup itself: the schema must describe what a reader can see. Marking up an FAQ that is not on the page is a guidelines violation and it can cost a whole site its rich results. The markup is a label, not a claim.

The things that stop extraction entirely

Any one of these is enough.

• The content is injected by JavaScript. Many crawlers do not run it.
• The page is behind a login or a form.
• The answer is only in an image or a diagram.
• The page has no dates, and a competitor's has last month's.
• The page is a PDF. Better than nothing, worse than HTML.

Check the first one with one command:

curl -s https://example.com/your-page | head -c 2000

If your content is not in that output, nothing else on this page matters.

What does not work

• Keyword density. Extraction does not count keywords.
• Word count for its own sake. A 3,000 word page that never states the answer loses to a 600 word page that does.
• Writing "in 2026" into every heading.
• Self-serving comparison pages. Discounted by models and by readers.
• Hidden text of any kind. It does nothing for extraction and it is a penalty.

The ten-minute pass

For any page you already have:

1. Find the question it answers. Make that the heading.
2. Write the 40 to 60 word answer. Put it directly under the heading.
3. Turn the longest comparison into a table.
4. Add one specific number with a date.
5. Put a visible published and updated date on the page.
6. Check the content appears in curl.

Six steps, ten minutes, and it works on every page you have already written.