Free tool · extraction
Which of your paragraphs survive being quoted?
An assistant lifts a passage and leaves the heading behind. This reads your page section by section and tells you which ones still say something once that happens.
01The four rules
A quotable passage has to work with the page taken away from it.
- 01It has to name its own subject
- Extraction takes the paragraph and leaves the heading behind. A passage under 'What is answer readiness' that begins 'It is the practice of…' arrives at the model as a sentence about nothing, and gets dropped or misattributed.
- 02It has to answer in one block
- A model lifting a passage takes a contiguous run of text. An answer split across a paragraph, a list, and a second paragraph three screens down is an answer the model has to assemble, and assembling is where it decides to use somebody else instead.
- 03It has to be the right size
- Under about twenty-five words a passage is a label rather than an answer. Past a couple of hundred the answer is in there somewhere and the model has to hunt for it. Neither extreme is a rule of style; both change whether the text gets used.
- 04It has to stay on the question it was asked
- A section headed with one subject whose opening paragraph never mentions that subject reads fine in place and reads as unrelated once it is quoted. The heading was carrying the meaning, and the heading is the part that gets thrown away.
02Questions
- Does this use AI to judge my writing?
- No. Every verdict is a structural rule applied to the text: how long the passage is, whether it opens on a pronoun, whether it restates the subject of its heading, whether a paragraph exists under the heading at all. Run it twice and you get the same answer, which is not true of a model-scored check.
- Is a low score bad writing?
- Not necessarily. Prose written to be read in order can be excellent and still fail here, because reading in order is exactly the assumption extraction breaks. What this measures is whether the page also works in fragments.
- Why does it only read the first paragraph of each section?
- Because that is the paragraph most likely to be lifted. A model looking for an answer under a heading takes what is directly beneath it, so the first paragraph carries a disproportionate share of the work.
- It flagged a passage I think is fine.
- Read it aloud with the heading covered. If a listener could not tell what it was about, the flag is right. If they could, the rule has misfired, which happens on headings whose subject is implied rather than stated, and the finding can be ignored.
- Should every section pass?
- No. Navigation, transitions and closing sections have no answer in them and are not supposed to. What matters is that the sections carrying your actual claims are quotable, and those are the ones worth reading the report for.
- Does fixing this guarantee citations?
- No, and nothing does. Quotable passages are one of four preconditions, alongside crawler access, rendering and being identifiable as an entity. This check covers one of them and says so rather than scoring the other three by implication.
Next
Quotable content nobody can fetch is still invisible.
The full check scores all four preconditions together: crawler access, rendering, machine-readable structure and answer readiness, with every finding traced back to a line in your HTML.