September 7, 2026 · 6 min read

AI Detection for Teachers: What It Can and Can't Tell You

Thinking of running student work through an AI detector? Here's how these tools actually work, why false positives happen, and a fairer way to use them.

AI detectors are now part of most teachers' lives, and they're widely misunderstood on both sides of the desk. Used well, they're a prompt for a conversation. Used as proof, they cause real harm to honest students. Here's what they actually measure and how to use one fairly.

How these tools work

An AI detector doesn't compare text against a database of generated writing. It's a statistical model that estimates how predictable the text is — how uniform the sentence lengths are, how often the most expected word appears, how little variation there is. Generated text tends to be smoother and more even than human writing, and that smoothness is what gets scored.

Why false positives happen — and who they hit

• Non-native English writers often write in a plainer, more regular style, which scores as machine-like. This is a documented fairness problem.
• Students explicitly taught formulaic structures — topic sentence, three supports, restatement — produce exactly the pattern detectors react to.
• Neurodivergent students and anyone with a very consistent writing voice can score higher.
• Ordinary tools blur the line: grammar checkers, translation, and 'improve this sentence' features all nudge writing toward the smooth patterns detectors flag.

What a score is and isn't

A percentage is a probability estimate about text patterns, not evidence about a person. It cannot tell you who wrote something, and it cannot distinguish 'used AI to draft this' from 'writes in a very regular style'. Detector vendors themselves say the output requires human judgement. No score should ever be the sole basis for an academic-integrity decision.

A fairer way to use it

• Treat a high score as a reason to look more closely, never as a finding.
• Compare against the student's previous work — a sudden, dramatic change in voice is far more informative than any percentage.
• Ask about the process: outlines, drafts, notes, version history. A student who wrote the essay can talk about how they built it.
• Start from curiosity, not accusation. 'Walk me through how you approached this' surfaces the truth in either direction without damaging a student who did nothing wrong.
• Consider designing assessments that are harder to outsource — in-class writing, drafts, oral defences — rather than policing the output alone.

A free tool for a quick look

If you want to scan a passage yourself, Skrubly's AI Content Detector gives an estimate and highlights the specific passages that read as most AI-like — which is more useful than a single number, because it shows you what to actually discuss. It's free, requires no sign-up, and runs entirely in your browser, so student work isn't uploaded or stored anywhere.

Worth telling your students

Many students who wrote their own work are quietly terrified of being falsely flagged. Telling them how these tools work — and that you won't treat a score as proof — removes a lot of anxiety, and often does more for honesty than the detector itself.