How to write a citable passage.
A citable passage is a section of text that stays correct and comprehensible on its own even when torn out of its page.
Once the technical side of answer engine visibility — access, structure, schema — is solved, what remains is a writing job, and that is where the difference is made. Two sites can share identical infrastructure; what decides which one gets cited is how the sentences are built.
The rule fits in one sentence: a passage must stand when torn out of the page. When a model lifts a section into its answer, the section does not bring its surrounding context along. Every sentence that leans on context collapses the moment it is separated.
Back-references kill
The habit that damages citability most is the back-reference: “as we explained above,” “in the first step of this method,” “the problem just mentioned.” These give the reader flow, but quoted they mean nothing. A model hesitates to select a passage that cannot be understood without context, and usually moves to another source.
The solution is not to sacrifice flow but to replace back-references with short repetitions. Writing “the contrast ratio calculation” instead of “the method we described above” makes the sentence independent without tiring the reader. Repetition usually counts as a flaw in writing; where citability is concerned, it is a virtue.
- Bad: “This method works in most cases.” — which method?
- Good: “Tying the contrast ratio to a token-level test provides a lasting fix in most cases.”
- Bad: “The three rules we listed above are enough.” — which rules?
- Good: “Keyboard access, visible focus and alternative text — the three cover most of level A.”
- Bad: “So what should be done about this?” — a question carries no information.
- Good: “Answering the reduced-motion preference with a full alternative flow is more reliable than merely damping the animation.”
The first-sentence rule
The first sentence of a section determines that section’s chance of being cited out of all proportion. The model looks for a short statement correct on its own; the first sentence is the nearest candidate for the role, and usually the first place looked.
A good first sentence does three jobs. It answers the question the heading asks — answers, not repeats. It needs no outside context. And it cannot be misread when read alone. When all three hold, the sentence is both citable and a good opening for the reader.
There is a practical test: copy the section’s first sentence, paste it somewhere else, and read it. Does it still say something, or is it hanging in the air? If it hangs, rewrite it. The test takes ten seconds and can be applied to every section on a page.
Concreteness
Models prefer verifiable statements, and verifiability runs through concreteness. A passage carrying a number, a date, a measurement or a definition is cited visibly more often than one saying the same thing in general terms. The likely reason is cross-verification: a concrete claim can be checked against other sources; a vague one cannot.
This also decides what the content should say. “Performance matters” is not a passage; “the body text measures 18.15:1 contrast against the paper background, where the AAA threshold is 7:1” is. The second is measured, verifiable, and stands when quoted.
A measured sentence gets cited; a claimed sentence gets skipped.
Length
Cited passages are usually short — one or two sentences. Long paragraphs are rarely taken whole; the model either cuts a piece or skips entirely. This sets a practical bound for paragraph length: one idea, one paragraph, and the paragraph of reasonable size.
But brevity alone is not enough. A short, empty sentence — “this topic is important” — does not get cited. What is sought is brevity together with information density: a complete claim in few words.
One last reminder: none of this is a special trick for the model. Writing independently, concretely and briefly is already the definition of good writing. What changed in the answer engine era is that the cost and the payoff of this discipline both grew.
The heading and the passage
Cited passages are almost always found right under a heading, and this is no accident: the heading lets the model understand which question the passage answers. The heading defines the question; the first paragraph gives the answer. That pair is the basic unit of citable content.
The practical consequence is that headings should be conceived in the language of questions — they need not be phrased as questions, but they must correspond to one. “Implementation details” is not a heading; “How contrast is measured” is. The first answers no question; the second is the exact counterpart of one.
This also explains why the heading hierarchy serves three readers at once: a table of contents for the screen reader user, document structure for the search engine, passage boundaries for the answer engine. One discipline, three separate benefits.
Structured lists
Numbered and bulleted lists have an advantage over paragraphs for citability because they are already divided into units. A model can take a step list whole or pull a single item; the structure survives either way. This explains why lists are cited so often in “how to” content.
But the list has its own discipline: every item must be complete in itself. “Measure first” is not enough; “Compute the contrast ratio from the palette file — do not judge it by eye” is a complete item. A list of half-items looks tidy but says nothing when quoted.
There is a length bound too: a fifteen-item list does not get cited, it gets cut. Five to eight items is a range that both reads well and quotes well. If a longer list is needed, grouping under subheadings helps reader and model alike.
And finally, the limit of this discipline: trying to make everything citable can leave the text dry and robotic. A piece’s narrative flow, its examples and its tone have value too; text written for the reader always beats text written only for the machine. The right balance keeps the KEY sections independent and concrete while letting the connective tissue run free. Not every paragraph needs to be citable — it is enough that the answer to the page’s central question is.