Turnitin's detector learned mostly from older AI writing. Newer models write plainer, more varied text, so Turnitin can score it very low, sometimes 0%. Other detectors trained on newer models can still flag the same text.
Written by Adithyan, a student who builds PaperCheck. Last updated 16 October 2026.
Older ChatGPT text was easy to spot: words like moreover and furthermore, phrases like plays a crucial role, every sentence about the same length. Newer models mostly dropped those habits, so a lot of the old signals are gone.
In the 2026 study by Van Vlasselaer and others, fully AI-written papers from the newest models averaged 0% on Turnitin. In my own testing, reports that a GPTZero-style tool rated around 97% AI got under 2% from older open-source detectors.
More colleges use more than one tool now, and faculty read the work too. 0% on one detector isn't proof the writing is yours, and it says nothing about what other tools will show.
Not reliably. A 2026 study found it gave 0% to papers written fully by the newest models.
Only for Turnitin. Tools like Pangram or GPTZero may still flag text from newer models.
Check your paper: 5 free checks a month
PaperCheck is not affiliated with Turnitin, GPTZero or Pangram. Scores from any detector are estimates, not proof.