Can Turnitin detect rewritten AI text? test results inside

One of the most common questions I see is whether manually rewriting AI-generated text will fool Turnitin. I set up a controlled test to find out.

The Experiment

I took a 1,000-word ChatGPT essay on climate change and created five versions with increasing levels of human editing:

Version 1: Raw ChatGPT output (no changes)
Version 2: Fixed typos and minor word swaps (5 minutes of editing)
Version 3: Rewrote topic sentences and transitions (15 minutes)
Version 4: Restructured paragraphs and added personal examples (30 minutes)
Version 5: Kept only the ideas, rewrote everything from scratch (45 minutes)

Turnitin Results

Version 1: 97% AI detected
Version 2: 89% AI detected
Version 3: 62% AI detected
Version 4: 28% AI detected
Version 5: 4% AI detected

What This Tells Us

Light editing barely moves the needle. You need to fundamentally change the structure and add genuinely original content to get below detection thresholds. By the time you’ve done enough rewriting to reliably avoid detection (Version 4-5 level), you’ve essentially written the paper yourself.

The Irony

Version 5 took 45 minutes and produced a paper that was entirely in my own words. Writing the paper from scratch without AI would have taken maybe 60-90 minutes. So the time “saved” by using AI was minimal, and the stress of worrying about detection isn’t worth the 15-45 minutes you might save.

The Real Lesson

If you’re going to spend 30+ minutes rewriting AI text to avoid detection, you’re better off spending that time writing your own paper. At least then you learn something and have zero risk of academic consequences. The shortcut isn’t really a shortcut.