How to Detect AI-Generated Student Work (Honest Guide)
You suspect a student used AI to write their essay. What do you do? The answer is more nuanced than “run it through a detector.”
AI Detection Tools Are Unreliable
Let’s get this out of the way: GPTZero, Turnitin’s AI detector, and similar tools have significant false positive rates. Studies have shown they flag non-native English speakers’ writing as AI-generated at disproportionately high rates. They also miss AI writing that’s been lightly edited.
You should not use an AI detection score as the sole basis for an academic integrity accusation. Period.
What Actually Works
1. Know your students’ writing
The best AI detector is a teacher who reads their students’ work regularly. If a student who writes in short, simple sentences suddenly submits a paper with complex subordinate clauses and academic vocabulary — that’s a signal.
Keep early writing samples from each student. Compare voice, vocabulary level, sentence structure, and common errors. AI doesn’t make the same mistakes your students make.
2. Look for these red flags
Suspiciously perfect structure. AI writes in clean, predictable patterns: intro with thesis, body paragraphs with topic sentences, conclusion that restates the thesis. Real student writing is messier.
No personal voice. AI writing is competent but bland. It lacks the quirks, opinions, and specific experiences that make student writing human.
Vague examples. AI generates plausible-sounding but generic examples. “For instance, many studies have shown…” without citing a specific study. Students who did the reading cite specific details.
Consistent quality throughout. Real student writing has strong parts and weak parts. AI writing is uniformly mediocre — no brilliant insights, no obvious struggles.
Vocabulary mismatch. A 7th grader using “furthermore,” “paradigm,” and “multifaceted” in a single paragraph didn’t write that paragraph.
3. The conversation test
If you suspect AI use, don’t accuse — ask questions about the writing:
- “Tell me about your writing process for this essay.”
- “What was the hardest part to write?”
- “Can you explain what you meant in this paragraph?”
- “Where did you find this information?”
A student who wrote the essay can answer these easily. A student who submitted AI output will struggle to explain their own “thinking.”
4. Process-based assessment
The most effective anti-AI strategy isn’t detection — it’s designing assignments that require visible process:
- Require drafts. First draft, peer review, revision, final draft. AI can generate a final product, but it can’t fake a revision process.
- In-class writing. Even 15 minutes of in-class writing gives you a baseline for each student’s actual ability.
- Annotated sources. Require students to annotate their sources and explain how each one informed their argument.
- Personal connection. “How does this topic connect to your own experience?” is hard to fake with AI.
What to Do When You Suspect AI Use
- Don’t accuse publicly. Private conversation only.
- Ask about the process (see conversation test above).
- Compare to known work. Pull up their previous writing.
- Focus on learning, not punishment. “I want to make sure you’re getting the learning from this assignment. Can you walk me through your thinking?”
- Follow your school’s policy. If your school has an AI policy, follow it. If it doesn’t, advocate for one.
The Bigger Picture
The goal isn’t to catch students cheating. The goal is to design learning experiences where AI is either a tool (used transparently) or irrelevant (because the assignment requires human thinking).
An essay prompt like “Summarize the causes of World War I” is trivially AI-able. A prompt like “Interview a family member about how a historical event affected your family, and connect their story to what we learned in class” is not.
The best response to AI in education isn’t better detection — it’s better assignment design.