AI-generated work has patterns. Not always. Not visibly to the naked eye. But patterns nonetheless. You'll see them if you know what to look for. And once you do, you have a choice: assume the worst and accuse, or assume confusion and investigate.
Most of the time, the student doesn't think they did anything wrong. They thought using AI was allowed. Or they thought they were just "getting help." Or they didn't understand the assignment. Accusation makes them defensive. Curiosity makes them honest.
The pattern to watch for
AI-written text tends to:
- Open with a thesis statement immediately, like they're answering an essay prompt, even if the prompt didn't ask for one.
- Sound polished but slightly generic in a way that's hard to pin down. Not quite their voice.
- Transition smoothly between ideas in ways that feel a little too buttoned-up. Students learning to write usually stumble a little.
- Use sophisticated vocabulary a bit too evenly, like every paragraph has one "elevated" word, not scattered naturally.
- Never sound uncertain. Students backtrack. They change direction. They say things like "I think" or "It seems like." AI sounds confident about everything.
- Include phrases that are technically true but oddly placed. Like mentioning a nuance that wasn't asked for but the AI threw in anyway.
None of these things *proves* AI. But together, they add up to a "wait, let me ask" moment.
How to know for sure
Don't ask the student outright. Ask them *live*, in your presence:
"Tell me about this section of your work. Explain the thinking here."
If they can explain it easily, defend their choices, modify it on the spot when you push back—they understand the material. They probably just got help with the writing.
If they're vague, or they repeat what they wrote instead of explaining it, or they can't defend a statement when you ask why they said it that way—something's off.
You can also ask them to point out the part they're least sure about. Real learners always know that part. AI-generated work makes students uncertain about *everything*, because it's not actually theirs.
The conversation that isn't accusatory
Here's the move:
"Walk me through this work. I want to understand your thinking here because some of this sounds polished in a way that's different from your usual writing. Help me understand where your thoughts end and where you got help."
Notice: "Got help" isn't the same as "cheated." It opens a conversation, not a trial.
Most of the time, they'll tell you. "I asked ChatGPT to help me start it" or "I pasted in my ideas and asked it to make them more formal." Now you know.
Then: "OK, so you had the ideas. That's the learning. The phrasing came from the AI. For next assignment, I want to see your ideas in your own words, even if they're rough. Your thinking is what I'm assessing, not your ability to sound polished."
Most students nod. They didn't realize the distinction.
The policy move
The conversation means nothing if your syllabus doesn't say clearly:
- What AI use is allowed for this assignment
- What it's not allowed for
- What happens if they cross the line
"You can use AI to brainstorm and organize ideas. You cannot use it to generate full sentences for the assignment. If I see work that isn't yours, we'll meet and talk about it first before any grade impact."
Clear beats vague. Students want to know the rules. They don't want to guess and fail.
Where this gets clearer
Our [grading rubric tool](/grading-rubric) helps you build a rubric that explicitly includes "demonstrates independent thinking" as a criterion, and our [student feedback](/student-feedback) guide gives you language for these conversations that's firm but not accusatory.
The hardest part
The hard part isn't spotting the work. It's believing that most of the time, the student thought what they were doing was OK. They're not trying to commit fraud. They're testing boundaries on something they don't fully understand yet.
Treat them that way. Get curious. Have the conversation. Most of the time, they learn from it.
