You’ve been grading discussion posts for half an hour, and they’re all fine. Clean paragraphs, the word “delve” on repeat, an “overall” sentence to wrap things up, a source or two that don’t exist. Then one stops you: it’s a little rough, a little uneven, and it sounds like a person wrote it. That’s what stands out now: not the AI; the human.
Notice what that moment tells you. We can recognize AI all day, and it changes nothing, because recognizing it was never the hard part. The hard part is the question underlying every submission: did this student actually learn anything? A polished paper used to answer that for us. Now it doesn’t, and online (where the writing is usually all we get) there’s little else to go on. We read the work and honestly can’t tell whether the understanding is the student’s or the model’s. GenAI has become a confounding variable that makes it impossible to assess what a student does and does not know. We didn’t just lose a way to grade; we lost a way to see who needs help.
The banning dead end
Still, the reflex is to police students’ AI use: detect it, ban it, restore order. But this strategy is a dead end. The tells you noticed won’t hold up as proof, and the tools sold to provide proof don’t work. Tested against more than a dozen AI detectors, researchers found them unreliable, often passing AI text off as human and easily fooled by a few minutes of editing (Weber-Wulff et al., 2023). They’re biased too; flagging non-native English speakers far more often and punishing students for plain prose rather than dishonesty (Liang et al., 2023). OpenAI shut down its own detector for being inaccurate. And even if one worked perfectly, it would only tell you which tool a student used, never whether they learned anything. Policing AI just turns teaching into surveillance: a case you can’t win and students who treat assignments like a trap.
Ask a different question
Quit asking “did they use AI?” Start asking “can I tell they learned this?” Three moves get you there.
- Tell them why and draw the line. “Use your own words” is noise unless students know why it matters: the writing is the thinking and handing it off hands off the learning. So be specific about where the tool helps and where it doesn’t. Outlining or smoothing a clunky sentence? Fine. Producing the analysis that was the whole assignment? That’s the part you needed them to do. Vague bans teach students to be sneakier; a clear line and purpose give them a reason to play it straight.
- Grade the thinking, not the polish. Students spend effort where the points are. If the whole grade rides on a polished final product, you’ve announced that polish is the goal, and polish is the one thing AI hands out for free. Care about reasoning? Put the points on it, and spread them across the work: an outline, a draft, the choice they had to defend, a peer response, the ninety-second video explaining why they did it that way. Graded in stages, the thinking shows up across the term instead of in one tidy artifact at the end, and the learning becomes visible by design.
- Verify the learning, not the tool. This is the one that matters, and where most of the energy should go. When a submission won’t tell you what a student knows, stop trying to establish what they did and start confirming what they learned. That’s learning verification, and the reframe changes everything. Your suspicion isn’t a charge to prove; it’s a flag that you’re missing the information the assignment was supposed to give you. So go get it directly and ask the student to show you the understanding. Not an accusation, not a misconduct case, just the most ordinary thing a teacher does: checking whether the learning happened. The impossible question (how do I prove AI use?) becomes a simple one (what’s another way for this student to show me they get it?), and there are plenty of answers, even online.
How to verify learning online
None of these is a misconduct process. Each one just gives a student another way to show you the learning, and online you have more of them than you’d think:
- Have them explain it in their own words. A two-minute video or a voice memo: what’s your main point, and how did you get there? A student who did the thinking can do it on the spot. You’re listening for whether the understanding is theirs.
- Ask them to walk you through a source. Instead of hunting for fabricated citations, pick one reference and ask what it argues and why they chose it. If they engaged with the reading, it shows in seconds. If they didn’t, you’ve found the gap you needed to teach into, which was the point all along.
- Ask for a short reflection. A quick video of how their thinking changed and what tripped them up. Reflection is where understanding actually lives, and it’s the hardest thing to fake when the understanding isn’t there.
- Get on a quick call. Five minutes, curious instead of prosecutorial: walk me through this part; what would you change with another week? You’ll hear the depth of their understanding fast, and they get a fair shot at showing it.
None of this is really about grades. It’s about getting back the thing AI quietly took from us: a clear view of our students. Verifying learning is how you catch the one who’s drowning behind fluent paragraphs, and how you find the one who’s genuinely wrestling with the ideas and deserves to be pushed further. That view matters in any classroom, and online the need is intensified. AI didn’t so much break assessment as expose it, pulling the cover off an assumption we’d leaned on for years: that a clean paper meant a lesson learned. It often didn’t. The real work now is to stop guessing what our students know and start seeing it, which is what teaching them was always supposed to be.
References
Liang, W., Yuksekgonul, M., Mao, Y., Wu, E., & Zou, J. (2023). GPT detectors are biased against non-native English writers. Patterns, 4(7), Article 100779. https://doi.org/10.1016/j.patter.2023.100779
Weber-Wulff, D., Anohina-Naumeca, A., Bjelobaba, S., Foltýnek, T., Guerrero-Dib, J., Popoola, O., Šigut, P., & Waddington, L. (2023). Testing of detection tools for AI-generated text. International Journal for Educational Integrity, 19(1), Article 26. https://doi.org/10.1007/s40979-023-00146-z
AI Disclosure
The author used a generative AI (Claude) assistant to help draft and refine this article. The argument, structure, revisions, and final wording are the author’s own, and the author takes full responsibility for its content.
B. Jean Mandernach, Ph.D., is Executive Director of the Center for Innovation in Research on Teaching and the Center for Educational Technology and Learning Advancement at Grand Canyon University. Her research examines online learning, assessment, faculty development, analytics, and artificial intelligence, emphasizing effective, engaging, and sustainable instructional practices for students.
