Leave the unflagged prose alone — rewrite only what the report painted

You opened the detector report. Half the page is quiet. A few continuous spans are painted red. The instinct is to "fix" the whole document so the score goes down. That instinct is usually wrong.

This note is about a narrower habit: leave the unflagged prose alone, and spend rewrite time only on what the report actually marked.

What the paint is telling you

Most academic AI-writing reports are span-level, not page-level. The percentage at the top is a summary of qualifying text that matched the model. The useful part of the file is the highlight map: which continuous prose blocks were flagged, and which were not.

Unflagged paragraphs are not a free pass from your instructor. They are also not a reason to rewrite them "just in case." If you rewrite clean spans, you introduce new wording into regions the report never asked you to touch — and you burn time you needed for the painted ones.

A practical edit order

1. Import the report next to the DOCX (or PDF export of the same paper).
2. List only the flagged continuous prose spans — ignore headers, captions, and quiet paragraphs for the first pass.
3. Rewrite those spans in place. Keep citations, numbers, and equation blocks frozen unless the highlight literally covers them.
4. Leave unflagged prose alone until a later recheck, if you even need one.

That order matches how the report is built: the paint is the worklist. Everything else is context.

What this does not claim

No workflow here promises a particular detector outcome. Detectors update; schools disagree; short documents can swing hard. The claim is narrower: when you do edit, start with the marked continuous prose and do not casually rewrite the quiet parts.

If you want a document-level pass that follows that same rule — import the report, touch flagged spans first, keep layout and citations stable — see humanpen.net.

— HumanPen Team