Charles Li / HumanPen
When an AI writing percentage appears on Moodle, Canvas, or Blackboard, many students in India and Southeast Asia treat it like a grade for the entire submission: cover page, tables, reference list, appendix protocols, figure captions, and all. That is usually the wrong mental model. Most institutional pipelines estimate AI writing patterns on qualifying continuous prose. Large stretches of your file may never enter the numerator or the denominator the way your essay body does.
This post is about partial-coverage literacy—how to read a percentage when tables and appendices are effectively unscored. It does not promise any score, and it is not a detection-evasion guide.
Reference lists, numbered tables, short captions, bullet-heavy appendices, and formula blocks often contribute little or no continuous prose under a given tool’s rules. The card may still show a confident-looking percentage for the introduction and literature review while your methods appendix sits gray, blank, or simply absent from the highlight map.
That asymmetry produces two opposite panics in WhatsApp groups:
Neither claim follows. The percentage answers a narrower question: among the text the tool treated as scorable continuous prose in this upload, how much looked model-like under this estimator?
| Often in continuous-prose scoring | Often outside or lightly scored | What to do |
|---|---|---|
| Introduction, lit review, discussion paragraphs | Standalone tables / spreadsheet dumps | Check the highlight map, not the badge |
| Narrative methods paragraphs | Bullet protocols, checklists | Ask whether your school wants prose or appendices revised |
| Long figure explanations written as paragraphs | One-line captions | Do not “pad” captions to chase a number |
| Abstract (when long enough) | Reference list / bibliography | Keep citations frozen unless factually wrong |
Product details differ. Your job is not to reverse-engineer a vendor. Your job is to open the full report and see which pages actually carry highlights.
If the card also shows `\*%`, a gray “unprocessed” strip, or a confidence band, the headline number may be both partial (not whole-file) and conditional (not finished). Refresh, download the official export, and read footnotes before you rewrite. Students in Bengaluru, Singapore, and Jakarta lose hours arguing about a badge that was still processing.
A low similarity score does not cancel an AI writing indicator. An AI writing indicator does not prove plagiarism. Partial coverage makes both instruments easier to misread, because a classmate’s appendix-heavy PDF is not a fair comparison for your prose-heavy DOCX.
Bring: full report (with footnotes), the exact attempt file, a one-page map of highlighted vs gray vs never-shown sections, and the module clause on AI writing tools. Ask clearly: “For this assignment, should revision focus on flagged continuous prose only?” That question is more useful than “Is 24% bad?”
If revision is required, confirm the file identity in writing, then edit flagged passages first. Freeze tables, citations, and checked facts unless they are wrong.
A document-level rewrite workflow—import the report, revise only flagged text, keep layout—is available at HumanPen, with no promise about the next percentage.