~/answers

Are AI detectors accurate?

Not accurate enough to prove that someone used AI. Several universities have switched AI detection off, citing false positives and bias against writers whose first language is not English.

What universities found

• Vanderbilt disabled Turnitin’s AI detector in August 2023. Even a 1% false-positive rate, the figure Turnitin gave, means many students wrongly flagged each year, and it raised concerns about bias against non-native writers (Vanderbilt).
• The University of Waterloo stopped using Turnitin’s AI detection in September 2025, calling the tools unreliable and pointing instructors to assessment design, AI literacy and disclosure instead (Waterloo).
• A Stanford study found that seven detectors flagged 61% of TOEFL essays by non-native English writers as AI-written, while doing far better on essays by US students (Stanford HAI).

Why detectors struggle

A detector looks at how predictable the words are. Clear, simple writing, writing in a second language and heavily edited text can all look predictable, so honest work gets flagged. Light rewording of AI text can go the other way.

What works better

Looking at how a piece was written: drafts, version history and a conversation about the work. In Snowkit Write, students can keep a private writing record and share it with an AI-use statement. It is not a detector. It shows what happened while they wrote, and the student decides whether to share it.