AI & academic integrity
AI detectors are guessing. Your students deserve better.
Every detector hands you a probability and calls it a verdict. The false positives land on the honest and the non-native, the appeals land on your desk, and the cheaters keep slipping through. There’s a better frame: prove authorship instead of hunting for it.
The problem with detection
Three reasons a detector score can’t be trusted.
False positives fall on the innocent
A widely cited 2023 Stanford study found detectors flagged the majority of essays by non-native English speakers as AI-generated, while rarely misjudging native writers. The students least able to contest an accusation are the ones most likely to face one.
A percentage is not evidence
A detector hands you “37% AI” with no way to see how it decided. You can’t show it to a student, you can’t defend it in a hearing, and you can’t reproduce it. It shifts the burden of proof onto you — and then onto them.
It’s an arms race you lose
Every detector is trained to spot last month’s models, and a quick paraphrase slips past most of them. You end up policing tooling instead of teaching, and the students willing to cheat are the ones best equipped to evade you.
The alternative
Prove it, don’t detect it.
When students write in Folio, the work carries its own evidence. Each submission mints an Integrity Certificate built from the student’s own process — not a probability inferred after the fact:
- How the writing took shape, over time
- The sources they actually cited, verified
- Any AI assistance they disclosed, in the open
Honest students are protected because the record speaks for them. And if a grade is ever challenged, you point to evidence you can show — not a number you have to defend.
thesis-chapter-3.docx
Version timeline
Common questions
Are AI detectors accurate?
Not reliably. Independent testing has repeatedly shown meaningful false-positive and false-negative rates, and even the vendors caution against using a score as sole evidence of misconduct. A probability is not proof.
Can Turnitin or GPTZero be wrong?
Yes. Both can flag human-written work as AI and miss AI-assisted work, and their scores can shift on lightly edited text. Neither is designed to be conclusive evidence in an academic-integrity case.
Why do detectors flag non-native English writers?
Detectors often equate simpler, more predictable phrasing with machine generation — which is exactly how many fluent second-language writers compose. The result is a documented bias against non-native speakers.
What should I use instead of a detector?
Capture proof of authorship as the work is written, rather than guessing after the fact. Folio’s Integrity Certificate records how a piece took shape, the sources actually cited, and any AI use the student disclosed — evidence you can show, not a number you have to defend.
Prefer the full comparison? See Folio vs the campus LMS or the deeper proof-over-detection case.
Teach the way you meant to.
Free for instructors, no permission required. Bring one class over this week and see the difference proof makes.