Citable Passage

A citable passage is a block of prose that survives being quoted with zero surrounding context. It names its own subject, answers its question in the first sentence, and carries the evidence for its claim inside the same block.

The term is a property, not a size. That distinction is the whole reason this entry exists: nearly every guide in this niche defines the idea by a word count, and the count is the part we cannot source.

Why is a citable passage defined by context, not length?

Because retrieval operates on passages, and a passage arrives at the engine without the page around it. Documents are split into chunks before anything is searched, those chunks are scored against a query, and the winning chunk is what an answer gets composed from — the mechanics are at chunking and passage retrieval.

That pipeline is indifferent to how long your paragraph is. It cares only about whether the piece it carried away still means anything. A 30-word sentence that names its subject is citable; a 300-word section that opens with "this approach" is not.

What breaks a citable passage?

Five dependencies, in our experience of auditing our own pages. Each one is invisible while you read the page top to bottom, and fatal the moment the block travels alone.

The test is mechanical and takes ten seconds: cover the rest of the page and read the block. If you have to look up to understand it, so would a reader who met it inside somebody else's answer.

How does a contract enforce the property?

By making a fixed number of units on every page carry it, rather than trusting the discipline to hold across hundreds of pages. This library's contract enforces 2 such units — an answer block directly under the H1, and 3 to 6 standalone takeaways — on every page it publishes, and a build gate checks both before the site compiles [our data]. The same contract shape runs on our other production builds; the numeric bounds above are this library's own, not a fleet-wide specification [our data].

The sizing of the answer block is a separate decision with its own reasoning, including the fact that no engine publishes a preferred length: that argument is at how long a direct answer should be. Where each unit sits, and why the page order inverts the usual essay shape, is at how to structure a page so an LLM can quote it.

Enforcement matters more than the specific rule. A convention that lives in a style guide decays; a convention checked before the site compiles either holds or stops the build.

Does a citable passage get cited?

No. Selection is competitive against everything an engine has indexed, and nobody controls it — Google's own AI features documentation puts eligibility at being indexed and snippet-eligible, and eligibility is the floor rather than an outcome.

What the property buys is narrower and worth being precise about: when an engine does reach for one of your passages, the passage works. The closest thing to supporting evidence is the Princeton GEO study (KDD 2024), which measured benchmark visibility gains up to about 40% from passage-level edits that added citations, quotations, and statistics — a lab result about passage content, on the paper's own corpus, not a promise about anyone's page.

Against our own interest: this property is the cheapest thing in the discipline to copy. Anyone can write self-contained blocks by tomorrow afternoon, and if the sentence inside the block says nothing nobody else can say, a perfectly-shaped passage earns nothing. Structure is the container; the evidence is the content — which is the argument running through our operator's guide to generative engine optimization.