A quotable paragraph is 25 to 110 words and never opens with “This”

A citation is one passage, alone, on someone else’s screen. Not your page. Not your layout. One paragraph, lifted out and set down somewhere it has to make sense without anything around it.

If it cannot survive that trip, the engine summarises instead. A summary is a paraphrase, and a paraphrase carries no link back to you. You supplied the answer and someone else got the credit, which is the specific failure worth engineering against.

What is citation readiness?

Citation readiness is whether a page is built so a machine can extract a passage and attribute it. It is not the same thing as quality. A thoughtful essay with long, interlocking paragraphs can be nearly impossible to quote, and a plain page of short factual claims can be trivially quotable while saying much less.

What follows are the properties we check, and why each one decides whether a passage travels.

Passages that survive being lifted

The single most common failure is the dependent opening. A paragraph that begins This means that, These results show, It follows that or They typically refers to something the reader of the quote cannot see.

An engine has two options with a paragraph like that. It can quote it and produce a fragment that makes no sense, or it can resolve the reference by rewriting the passage in its own words. It picks the second one, and the rewrite is where your link goes missing.

The fix is mechanical: name the subject in the first sentence of every paragraph you would be happy to be quoted on. Not every paragraph, and the difference matters. Prose where every sentence restates its subject reads like a legal document.

Length works the same way. Below about 25 words there is rarely a full claim present. Above about 110, engines tend to compress rather than quote. That band is not a rule anybody published, it is where lifted passages tend to fall, and treating it as a target rather than a constraint produces exactly the choppy prose worth avoiding.

Shapes that come out of a page in one piece

Some formats extract cleanly and some do not. A list is already segmented, so an engine can take three items and leave the rest. A table has its own row boundaries. A question used as a heading, with its answer in the paragraph directly beneath, is a matched pair that can be lifted whole.

Continuous prose has none of those boundaries, so a machine has to invent them. Every invented boundary is a chance to paraphrase.

This is the most actionable finding in the whole area, and it costs nothing: take the three or four questions your page actually answers, make them h2 headings phrased as questions, and put the answer in the first paragraph under each. You have not changed a word of the substance.

Claims specific enough to be worth quoting

Engines quote what is specific. A page saying a process is significantly faster gives an answer nothing to carry. A page saying it takes four minutes instead of forty gives it a sentence.

The same applies to hedging. Some experts suggest and results may vary are unquotable by construction, because there is no claim in them to attribute. This is worth separating from honesty: stating a real uncertainty precisely is specific, and it quotes well.

Who said it, and when

Two pieces of metadata do most of the attribution work, and most sites have neither.

An author. A citation is a claim plus a source. If a page carries no author, no organisation and no schema markup naming either, the engine can still use your claim. It simply has nothing to credit.

A date. An assistant weighing whether to rely on a claim wants to know how old it is. An undated page is not evidence about now, it is evidence about some unspecified time, which is a weaker thing to rest an answer on. Publishing dates belong in article:published_time or a <time datetime> element, not only in rendered text.

Both of these are structured data problems as much as editorial ones, which is the subject of the schema types that actually get read.

Where your own claims come from

A page that cites nothing asks to be taken on faith. That is the position an engine is least willing to grant a source it does not recognise, and the fix is unglamorous: link to the standards, papers, docs and data you are drawing on.

There is a second effect here that has nothing to do with trust. Citing others is how a page joins a topic. A page with no outbound links sits outside every conversation it is trying to enter.

The limitation worth stating plainly

No engine has published its extraction rules. Everything above is inferred from what gets quoted and what gets paraphrased, across a lot of pages, and it is a description of observed behaviour rather than a specification. It will drift.

It is also not a ranking system. A page can be perfectly citable and never cited, because nothing here makes an engine want to answer with your page rather than someone else’s. Citation readiness removes a set of mechanical reasons you get passed over. It does not supply a reason to be chosen.

The honest framing is that this is the cheap half of the work. It is worth doing first because it is measurable and finite. What follows it, having something to say that nobody else has said, is neither.