You revised the introduction at midnight. You swapped two paragraphs in the discussion. You fixed a citation. In the morning the LMS still shows the same AI writing percentage—or a new number that does not match the file open on your laptop.
That gap is usually not a moral cliff. It is a version mismatch: the classifier ran on a specific upload at a specific time, and the document you are staring at now is not that upload.
This note is report literacy for that mismatch. It does not promise what the next percentage will be.
The short answer
An AI writing percentage is attached to a submitted file version, not to the living draft in your editor. If you changed prose after the report was generated—or if the LMS re-ran an older copy—the number on screen and the words on your screen can honestly disagree. Confirm which file produced the highlights before you rewrite or argue.
Ordinary ways drafts and reports diverge
Local edits that never left your disk
Multiple attempts; the open report is an earlier one
A downloaded AI Writing Report PDF generated hours before your last save
A new Word file with the same title (different object downstream)
Edits concentrated in lists, tables, or references that may sit outside qualifying prose
None of that requires a cheating narrative. All of it produces confusing screenshots.
What documentation pushes you toward
Public Turnitin-facing materials describe AI writing detection as a prediction over qualifying continuous prose in the text it was given—not as a live sync with every keystroke. Similarity matching remains separate. Authorship-related signals, where shown, are another dial again.
The denominator is the scored upload. Bibliography-only fixes may barely move the AI indicator.
Highlights are version-bound. Citing Tuesday’s highlight map against Wednesday’s draft mixes two exhibits.
Short qualifying runs stay blunt. Later polishing does not retrofit nuance into an old report PDF.
Display bands are not personality tests.*% on one version and a visible low-twenties figure on another can be two uploads meeting a display rule—not a confession arc.
Five-minute sync check
Match filename, word count, and submission timestamp to the draft you think you are defending.
Open the full AI Writing Report—not only the badge—and note highlights on that version.
List edits made after that timestamp in boring specifics.
Separate three stories: similarity overlap ≠ AI prose estimate ≠ authorship consistency.
Re-upload when you need the next report to match the draft, or revise against the report you actually hold.
How to describe the mismatch without sounding evasive
“This PDF is from submission time X. My current draft has these three post-report edits.”
“The highlights refer to paragraphs that no longer exist in draft Y.”
“I am asking about the AI writing report on upload X—not mixing similarity unless you open that dial.”
Avoid promising the next percentage. Avoid treating a local save as proof the LMS scored it.
Honest boundaries
Cross-tool numbers are not interchangeable versions of the same truth.
No public forever-list of “safe” rewriters survives detector updates.
Intentionally adding typos is not version control.
Local academic policy still decides process.
If revision is already decided
Import the Turnitin or iThenticate AI report that matches the upload you care about, rewrite only the marked continuous prose, and leave citations, tables, and formatting alone unless they are wrong. That document-level workflow is what HumanPen is built for—DOCX, PPTX, and TXT—without promising the next dashboard digit.
Takeaways
An AI percentage scores a submitted version, not the live file in your editor.
Confirm timestamp, highlights, and post-report edits before you panic-rewrite.
Keep similarity, AI writing, and authorship as separate questions.
Re-upload when you need the next report to match the draft you will defend.
Share the version story; do not share a lone digit as proof the rewrite “failed.”