Start writing

Guides

How to tell if a student used AI, without a detector

Last updated 24 October 2026

You usually cannot tell from the finished essay, and neither can a detector. What settles it is the work around the essay: the drafts, the history, the notes, and the student’s own account of it. Ask for those, talk to the student, and set the next assignment so that its process is part of what gets handed in.

Why not just run it through a detector?

Because a score is an estimate, and the people who study and sell detectors say so. MIT Sloan’s teaching guide is titled AI Detectors Don’t Work. Here’s What to Do Instead, and says the software has high error rates that “can lead instructors to falsely accuse students of misconduct.” It notes that across seven detectors, 61% of essays by non-native English speakers were wrongly flagged as AI, against about 5% of essays by native speakers.

Turnitin itself says its detector does not decide misconduct, and that because its false positive rate is not zero, instructors need to apply their “professional judgment”, their knowledge of the student, and the context of the assignment (Turnitin). Vanderbilt University turned Turnitin’s detector off in 2023 (Vanderbilt).

What can I look at instead?

  • The version history. A Google Doc, or a Word file on OneDrive, shows the essay growing over days. A teacher with edit access can step through it, and Draftback replays it like a film.
  • Drafts, notes and an outline, with their dates.
  • The sources: what the student read, and whether the essay’s claims are in them.
  • Their other work. Writing that is suddenly unlike everything a student has handed in before is a reason to ask, not a verdict.

None of these is proof either way. Together they are far better evidence than a percentage, and a student who wrote the essay can usually produce most of them.

How do I have the conversation?

Openly, and without the score as the opening line. Turnitin’s own advice is to decide your process before you need it, tell students what it is, and go in acknowledging that a false positive is possible, because not doing so makes the conversation “far more defensive and confrontational.”

Ask the student to talk you through the essay: how they found the argument, why a paragraph is where it is, what they cut. Ask for the history and the drafts. Write down what was said and shown, and remember that a nervous student is not a guilty one.

How do I set work that shows its own process?

MIT Sloan’s guide recommends clear rules about when and how AI may be used, examples of what is and is not allowed, and a mix of assessment rather than a retreat to handwritten exams, which raise their own equity problems. In practice:

  • Say what AI use is allowed in this assignment, with examples, and how to acknowledge it.
  • Ask for the process as part of the work: an outline, a draft, a short note on what changed.
  • Ask for the version history, or for the essay to be written somewhere that keeps one.
  • Keep a short conversation about the work as a normal part of marking, not a sign of suspicion.

Where does iwrote fit?

It is one way of setting work that carries its own record. Students write the assignment on iwrote, in the browser, and hand in a link with it. The link opens a certificate of how the piece was written, which you can check in your own browser against a file the student downloads. It is evidence about a writing session, not proof of authorship, and what it cannot establish is written down. Using iwrote with a class has the details.

Whatever you use, do not treat a student without a record as a suspect. Most writing is done without one, by people who wrote every word.

More guides