When I first started thinking seriously about AI in my classroom, my initial concern was the obvious one: students submitting AI-written work as their own. But that anxiety, I eventually realized, was pointing me in the wrong direction. The more interesting question wasn’t “how do I catch AI use?” It was “how do I design assignments where AI doing the work defeats the point?”
That shift changed how I think about everything.
The Shape, Without the Substance
There’s a phrase I keep coming back to: AI can produce the shape of learning without the substance. It can write an essay that hits all the structural markers — thesis, evidence, analysis, conclusion — while missing everything that makes writing genuinely educational. The struggle. The revision. The moment a student has to decide what they actually think.
This is the problem we’re not discussing honestly enough. When we talk about “AI-proof” assignments, we usually mean assignments that are hard to complete with AI. But that’s the wrong frame. The better question is: what does the cognitive work look like, and does AI do it instead of the student?
If the answer is yes, the assignment probably needs rethinking — not because AI cheating is rampant, but because the assignment was probably low-value even before AI showed up.
Designing So the Thinking Happens Anyway
Here’s what I’ve learned works: design for the process, not just the product.
A few approaches that have held up:
Make the thinking visible. Instead of just collecting a final essay, collect the steps — the initial notes, the question the student decided to pursue, the moment they changed their mind, the sentence they had to cut because it wasn’t actually true. AI can produce an essay, but it can’t produce a genuine record of a specific student’s thinking process.
Assign the conversation, not the output. I’ve shifted some of my assessments toward discussions, Socratic seminars, and brief oral explanations of written work. If a student submits a written argument and can’t explain the third paragraph in their own words, that tells me something important. And it doesn’t require any detection software.
Use specificity as a scaffold. Generic prompts get generic answers — from students and from AI alike. “Analyze the causes of World War I” is a low-resistance prompt for a language model. “Based on the document packet we used in class Tuesday, identify the two factors you found most convincing and explain what changed your thinking” is not. Specificity tied to classroom experience naturally filters for actual engagement.
Let the messy draft be the assignment. Some of my best assessment moments come from asking students to submit a rough, unpolished version of their thinking — and then reflect on what they didn’t know yet. AI produces polished text by default. A genuinely rough first draft, full of half-formed ideas and honest confusion, is harder to fake.
What AI Is Actually Good For Here
None of this means AI has no role in the writing process. It can be a useful thinking partner — helping students brainstorm, identify logical gaps in their arguments, or understand what a counterargument might look like. The difference is who’s doing the thinking.
I tell my students something like this: using AI to do your thinking is like hiring someone to do your workout. You’re paying for the outcome but skipping the part that actually changes you.
That’s not a rule about academic integrity — it’s a statement about what school is for.
The Deeper Shift
The most useful reframe I’ve found is this: stop asking “is this assignment AI-proof?” and start asking “is this assignment worth doing at all?”
If a student completing the assignment with full effort learns something — about the subject, about their own thinking, about how to construct an argument — then the assignment has value regardless of what AI can do. If the assignment is essentially a formatting exercise, AI doing it isn’t the problem. The problem was always there.
That’s the work of “working with the grain” — designing so that the intellectual labor is the point, and AI assisting or even completing the surface task doesn’t undermine what students actually need to develop.
It requires more thought upfront. But it’s also more honest about what we’re actually trying to do when we give students things to write.
This post draws on ideas developed in The AI Doesn’t Know Your Students, particularly Chapter 8: “Working With the Grain.” The book is available now.
