You've just been hit with a Turnitin AI false positive: Turnitin's flagged your essay for AI writing, and you know you wrote it yourself. Now what?
Here's a step-by-step way to check your own work, understand what Turnitin's own numbers actually say about false positives, and put together an evidence pack before you sit down with a tutor or panel.
None of this is about arguing with a machine. It's about giving a human reviewer something solid to look at.
- 01Turnitin's own blog states a document-level false positive rate under 1% for documents flagged with over 20% AI writing, from an 800,000-document test.
- 02At sentence level, Turnitin's own figure is closer to 4%, and errors cluster near transitions between human and AI writing.
- 03A free WatermarkRemoverPro check, done in your browser, is one extra piece of evidence, not a replacement for your own drafts and version history.
- 04Appeals go better with a dated, exportable record than with an argument alone.
- 05Save everything before you need it, not after.
Step 1: Save Your Original Draft and Version History
Before you do anything else, find your original files: every saved draft, not just the final one you submitted. Most word processors keep version history automatically; Google Docs has it under 'File > Version history', and Word has it under 'File > Info'.
If you wrote in stages, that history is worth more than any single tool's verdict. It shows a document growing over days or weeks, in your own typing rhythm, with your own false starts still visible.
In Google Docs, open 'File > Version history > See version history' and you'll get a timestamped list of every autosave, often down to the minute, showing exactly when text was added, deleted or moved around. Named versions (right-click any entry and choose 'Name this version') are worth creating at major milestones, such as finishing a first draft or working through feedback, because they're easier to point to later than an anonymous autosave from late one evening. Word behaves differently: it relies on AutoSave being switched on with the file stored in OneDrive or SharePoint, and past versions then sit under 'File > Info > Version History', or under 'Restore previous versions' when you right-click the file in File Explorer. If AutoSave wasn't turned on, Word won't have kept anything beyond your last manual save, which is exactly why saving early drafts under separate file names, such as essay-draft-1.docx and essay-draft-2.docx, is worth doing as a habit rather than only after a flag has already landed.
Step 2: Read What Turnitin Actually Flagged
Turnitin doesn't just give you a single 'AI' stamp. Open the report and look at exactly which sentences or sections got flagged, not just the overall percentage.
This matters because Turnitin's own published figures show errors aren't spread evenly. Their sentence-level false positive rate sits around 4%, and they've reported that 54% of those false-positive sentences sit right next to genuine AI writing, at the seams where styles change. If your flagged sentences cluster near a heavily quoted or paraphrased section, that's worth noting.
Step 3: Run Your Essay Through a Free WatermarkRemoverPro Check
Head to the Check page and paste in the flagged essay, or the specific flagged sections if it's long. The test runs entirely in your browser, and nothing gets uploaded anywhere, which matters when the document is still under review.
The free check covers up to 1,500 words. If your essay is longer, a free account raises that to 5,000 words and 20 checks a month.
Step 4: Understand What the Result Does and Does Not Show
Read your result against what it actually tests: a specific, keyed statistical watermark, not a general 'sounds like AI' judgement. If no mark turns up, that's honest information, but it isn't proof of human authorship on its own, because marks are keyed and no vendor publishes its detection key.
If you never used an AI writing tool, there's a good chance nothing will be found under any key WatermarkRemoverPro tests, simply because there's no mark to find. Either way, treat the result as one line in your evidence pack, not the headline.
Step 5: Gather Supporting Evidence of Your Process
Beyond the writing tool itself, gather anything that shows your process: research notes, an outline, screenshots of a search history, emails to a tutor about the topic, or a reading list. Ordinary, boring evidence of ordinary, boring work.
If you used any AI tool at all, even for brainstorming or grammar checking, be upfront about it in your notes. Trying to hide legitimate assistance you did use tends to backfire worse than disclosing it plainly.
Step 6: Build a Short Evidence Pack
Put it together as one short document: original drafts with timestamps, your WatermarkRemoverPro result, the specific sentences Turnitin flagged, and a brief, calm note explaining your process.
A WatermarkRemoverPro Pro account can export a dated PDF evidence report with the signal strength, a confidence band, the keys tested, and a SHA-256 hash of the document, useful if you want a formal, exportable record rather than a screenshot.
Step 7: Request Your Appeal Meeting and Stay Factual
Contact your tutor, module leader or academic integrity office to request a meeting, and bring the pack rather than emailing a long argument first. A calm, evidence-led conversation tends to go further than a defensive one.
Stick to what you can show, not what you feel. 'Here's my draft history and here's what a watermark check found' lands better than 'it's not fair', even when it genuinely isn't.
If your first contact has to be an email rather than a face-to-face request, keep it short and factual rather than pleading. A workable structure is: state plainly that you've been flagged, attach your evidence pack, and ask two direct questions, specifically what the panel's own documented false positive rate is for the tool that flagged you, and what further evidence they'd find useful ahead of any meeting. Something close to 'could you tell me what false positive rate your institution has recorded for this detector, and let me know what supporting material would be most useful before we meet' does two things at once: it shows you've engaged with the numbers seriously, and it gives a reviewer a specific, answerable question rather than an open-ended complaint they have to interpret for themselves.
| Level | Turnitin's stated figure | Detail |
|---|---|---|
| Document-level | Under 1% | For documents with over 20% AI writing, from an 800,000-document test set |
| Sentence-level | Approximately 4% | More common at transitions between human and AI writing; 54% of false-positive sentences sit next to actual AI writing |
“The students who come out of an appeal meeting fine are almost never the ones with the best argument. They're the ones who brought their drafts.”
Common pitfalls
- Deleting or not keeping draft history because the essay's already submitted; do this before you ever need it, every time.
- Treating a WatermarkRemoverPro 'no mark found' result as a guaranteed clean bill of health rather than one data point.
- Sending an angry first email instead of requesting a calm meeting with evidence attached.
- Hiding legitimate AI-assisted brainstorming or grammar-checking out of fear it'll look worse; it rarely does, compared with the alternative.
A detected mark is not proof of authorship, and an absent mark is not proof of human authorship. WatermarkRemoverPro's on-device rewrite can reduce detectable evidence but cannot guarantee defeating a vendor's undisclosed watermark, on any tier.
On WatermarkRemoverPro