How to Write a Page That Actually Gets Quoted
A model choosing what to quote is looking for a passage that answers the question completely and survives being removed from everything around it. Most marketing copy fails that test for reasons that have nothing to do with quality.
Ask an assistant a specific operational question about a product category and watch which single sentence gets lifted into the answer. It is rarely the most polished sentence on the page — it is the first one that actually answers the question without requiring the rest of the paragraph to make sense.
Here are the six properties that decide it, roughly in order of how often they are missing.
1. The answer comes first
If a section is headed "How long does onboarding take", the first sentence should state a duration. Not context, not a caveat, not a sentence explaining that it depends.
This inverts a habit that good writing spent years building. Persuasive copy builds — context, tension, payoff. Extractable copy leads with the payoff and explains afterwards.
The reason is mechanical. A system scanning for an answer takes the passage that contains one. If yours arrives in paragraph six, it takes someone else's paragraph one.
A worked example: a section headed "How long does onboarding take?" that opens with "Onboarding is a collaborative process that adapts to your organization's unique needs" contains no answer at all — a model reading it moves on to the next candidate. The same section rewritten to open with "Onboarding takes two to three weeks for most teams" gives the model a sentence to lift on the first try.
When the honest answer really does depend on something, lead with the range and the variable, not the hedge. "Setup takes one to three weeks, depending on how many integrations you connect" is extractable. "It depends on a number of factors" is not — it is the same information with the useful part removed.
2. The paragraph stands alone
Read any paragraph with nothing above it. If it does not make sense, it will be quoted out of context or skipped entirely.
The most common cause is pronouns pointing backwards. "This means you should audit it quarterly" is meaningless in isolation — what is this, and what is it? Naming the subject again feels repetitive when you read the page top to bottom. It is what makes the paragraph portable.
A fast way to check this: copy a paragraph into a blank document with nothing above it. If a stranger reading only that paragraph would have to ask what "this" or "it" refers to, the paragraph fails the test. The fix is almost always to repeat the subject's name once more than feels natural while writing — repetitive to the person reading the whole page top to bottom, invisible to someone who only ever sees the one paragraph.
3. One idea per section
Sections covering three related points give a model three partial answers rather than one complete one. It will usually take a cleaner passage from a page that separated them.
If a heading contains "and", it is often two sections. "Pricing and Support" is almost always better split into "How much does it cost?" and "What support is included?" — two headings a reader or a model can match to two different questions, instead of one heading that answers neither cleanly.
4. Facts live in sentences, not layout
Comparison tables, spec grids and infographics are excellent for humans and poor for extraction. A model can quote a sentence. It struggles to quote a grid, and it cannot quote an image at all.
Keep the table — it earns its place for readers. Then restate the two or three facts that matter most in a plain sentence nearby. You are not duplicating content; you are making the same information available in a second format.
A spec table showing "Storage: 50GB" as a row is invisible to a model looking for a quotable sentence. A nearby line reading "The Pro plan includes 50GB of storage" is not — it is the same fact, restated in a shape a system can lift directly.
5. Specifics beat adjectives
"Setup is quick and easy" cannot be quoted, because it asserts nothing checkable. "Setup takes about ten minutes and requires no code" gets quoted constantly.
The test is simple: could a reader disagree with this sentence? If not, it carries no information. Every unfalsifiable claim on a page is a sentence a model has no reason to reuse.
This is the single biggest difference between pages that get cited and pages that do not, and it is a copywriting problem rather than a technical one.
6. Headings are questions
Answers are generated for questions, so pages organised as questions have a structural advantage.
Not "Pricing" but "How much does it cost?" Not "Implementation" but "How long does setup take?" Then answer immediately underneath.
This also produces better pages for humans, which is a rare case where the two audiences want the same thing.
Quotable vs not, side by side
Seeing all six properties applied to the same kind of sentence makes the pattern easier to spot on your own pages.
How to test whether it is working
Pick three to five operational questions your own page claims to answer — pricing, setup time, what is included, how support works. Ask an assistant each one and check whether the answer matches your wording or a competitor's.
If your product is not named, or the wording is clearly not yours, go back to the specific section that should have answered that question and run it through the six properties above. Nine times out of ten, one of them is the gap.
Common mistakes
Leading with the caveat instead of the range. "It depends on your setup" is a sentence with no answer in it, even when a completely honest range would fit in the same space.
Writing the claim from the company's point of view instead of the fact's. "We're proud of our fast onboarding" is not extractable. "Onboarding takes two weeks" is — and it says the same thing.
Answering five sub-questions in one FAQ entry. A single long answer covering pricing, setup, and support together gives a model no clean sentence to lift for any one of the three.
Restating a fact only inside an image or a table with no plain-text sentence nearby. If the fact only exists as a graphic, it does not exist for extraction purposes.
What this does not mean
It does not mean writing shorter. Depth is fine. Burying the answer is not. A two-thousand-word page that answers in sentence one is entirely compatible with all of this.
It does not mean writing for machines. Every property above makes a page easier for a person in a hurry too. If your copy reads worse after applying them, they have been applied badly.
It does not mean removing personality. Voice lives in how you explain, argue and choose examples. It does not live in withholding the answer for three paragraphs.
A ten-minute edit
Take your most important page and do this.
- Read only the first sentence of each section. Do those sentences alone answer the page's questions? If not, the answers are buried.
- Find every pronoun in a first sentence and replace it with the noun.
- Find every adjective making an unfalsifiable claim and replace it with a number or delete it.
- Rewrite each heading as the question a reader would ask.
- Where a fact only lives in a table or image, add one plain sentence nearby that states it in words.
That edit typically takes ten minutes and does more than a week of adding words.
