How to Design Assignments Students Can’t (and Won’t Want to) Cheat On With AI
The ethical alternative to AI detection isn’t a better detector — it’s a better assignment. A two-bar framework: close the cheating path by design, and make the work worth doing for its own sake.
Answer-First Capsule (AEO Summary)
How do I design assignments students can’t and won’t want to cheat on with AI? Run the two-bar test on every assignment. Bar one — can’t cheat on it: design tasks that structurally require what AI can’t produce — oral defenses, visible process (draft history, reflection), local context, in-class collaboration, and staged iterative submissions. The cheating path is closed by design, not by threat. Bar two — won’t want to cheat on it: make the work authentic (real audience, real purpose), personally relevant, and give students agency in the design; name the cognitive work it builds and where they’ll use it again. A student who sees the point and chose the question doesn’t outsource the work. Pair the two bars with an explicit four-tier AI-use policy (off-limits / editor / research assistant / co-tool) so every assignment names what AI is for. Secondary AI scores assignments for AI-resistance and suggests revisions using these principles — run the two-bar test before you assign, not after.
Section 1: The Reframe
If an Assignment Can Be Done by AI, Maybe It Shouldn’t Be Assigned
The detection arms race is unwinnable, and as we showed in our last post, the detectors are biased against the students least able to defend themselves. So stop fighting on that front. The real lever is upstream: if a take-home essay can be completed by ChatGPT in 30 seconds, the problem isn’t the student who uses ChatGPT — it’s the assignment that asks for work AI can do.
The ethical question shifts from “did they use AI?” (which you can’t reliably answer) to “is this task worth doing if AI can do it?” (which you can). And the answer to that question is a design problem with a design solution: the two-bar test.
An AI-proof assignment clears two bars: a student can’t cheat on it, and a student won’t want to. Miss either bar and the assignment fails — for different reasons.
Section 2: The Two-Bar Test
Two Bars. Both Must Clear.
A cheatable-but-engaging task still gets cheated on. An uncheatable-but-boring task still gets shortchanged — students disengage, do the minimum, or find other ways to offload the thinking. The two bars are independent. Both must clear.
Can't Cheat On It
The task structurally resists AI completion — it requires skills, artifacts, or context AI can't produce. Oral defense, in-class synthesis, process history, local data. The cheating path is closed by design, not by threat.
Won't Want To Cheat On It
The task is worth doing for its own sake — it produces something the student values, connects to their experience, or builds a skill they want. A student who sees the point doesn't outsource the work.
Section 3: Bar One — Can’t Cheat On It
Six Design Principles That Close the Cheating Path
The goal isn’t to threaten students out of cheating — it’s to structurally remove the cheating path. Each principle below makes the task require something AI can’t produce, by design.
Make the thinking audible
Oral defenses, Socratic seminars, recorded explanations, in-class presentations. The student has to produce the reasoning live, in their own voice, on the spot. AI can't speak for you in the room.
Make the process visible
Draft history, version control, annotated bibliographies, reflection on what changed and why. The work isn't the final artifact — it's the trail of thinking that produced it. AI can fake a product; it can't fake a process you watched happen.
Anchor it in local context
Apply the concept to your school, your community, your data, this week's news. AI has no access to the specific, situated context of your classroom — and that context is where the real learning lives.
Build in collaborative, in-class components
Fishbowl discussions, peer critique, group problem-solving with roles. The assessment is the participation, not a take-home artifact. You can't outsource being present.
Require metacognitive reflection
Ask the student to explain their own thinking: why they chose this approach, what they'd do differently, where they got stuck. A student who didn't do the thinking can't reflect on it — and reflection is itself the learning.
Use staged, iterative submissions
Proposal → outline → draft → revision → final, with feedback at each stage. Each stage is low-stakes and visible; the final product is the accumulation, not a last-minute paste. Cheating requires a single opaque submission — don't give students one.
Section 4: Bar Two — Won’t Want To Cheat On It
Four Principles That Make the Work Worth Doing
You can build an uncheatable task that students hate — and a task students hate is a task they’ll find a way to shortchange. The second bar is about the student’s relationship to the work. A student who sees the point and chose the question doesn’t outsource the answer.
Make it authentic
Real audience, real purpose, real stakes. A letter to the school board. A podcast for younger students. A solution to a problem in the building. When the work goes somewhere, students want to own it.
Make it personally relevant
Connect the task to the student's experience, identity, or goals. A history essay becomes a family migration map; a physics problem becomes the sport they play. Relevance is the antidote to the “why bother” that drives outsourcing.
Make the cognitive work the point — and name it
Tell students exactly what thinking this assignment builds and why that thinking matters to them. “This builds your ability to evaluate evidence — the skill that protects you from being manipulated.” Students who understand the transfer don't want to skip the reps.
Give students agency in the design
Let them choose the topic, the format, the angle within the learning goal. A student who chose the question is invested in the answer. Compliance work gets outsourced; chosen work gets owned.
The throughline: students cheat when the work feels like compliance — something done to them, for a grade, with no transfer they can see. Students don’t cheat on work that’s theirs. The second bar is about making the work theirs.
Section 5: The AI-Use Policy
Name What AI Is For — On Every Assignment
Ambiguity is where cheating hides. A blanket “no AI” policy is unenforceable and dishonest; a blanket “AI is fine” policy offloads the thinking. The honest move is to name, for every assignment, what role AI is allowed to play. Four tiers cover almost every case:
Tier 1: AI is off-limits
When the task IS the thinking — constructing an argument, deriving a proof, writing a first draft. The struggle is the point.
Tier 2: AI as editor/feedback
When the student has produced the generative work and AI is used to critique, suggest revisions, or catch errors. The student owns the draft; AI improves it.
Tier 3: AI as research assistant
When AI gathers and summarizes sources the student then evaluates, cites, and synthesizes. AI does the gathering; the student does the judging.
Tier 4: AI as co-tool
When the task is explicitly about using AI well — critiquing its output, prompt-engineering a solution, building with it. The AI use IS the learning.
The principle: the tier is a property of the assignment, not the student. A student who uses AI as an editor on a Tier 1 task is cheating; a student who refuses AI on a Tier 4 task is missing the point. Name the tier. Teach the tier. Enforce the tier.
Section 6: A Worked Example
From Cheatable to AI-Proof
Same learning goal (analyze causes of the Industrial Revolution). Same unit. One is a 30-second ChatGPT paste. The other clears both bars.
Before
Write a 5-page essay analyzing the causes of the Industrial Revolution, due in two weeks.
Fails both bars: fully AI-completable, and a generic prompt no student is invested in.
After
In class: choose one cause and defend it in a Socratic seminar (observed, graded). Take-home: write a 2-page position paper building on your seminar contribution, with draft history submitted. In class the following week: 5-minute oral defense of your strongest counterargument. Reflection: 1 page on how the seminar changed your thinking.
Clears both bars: in-class oral components (can’t outsource), draft history (visible process), personal reflection (metacognitive), and the seminar gives the take-home work a point the student owns.
Section 7: What to Do Monday Morning
A Teacher’s Action Checklist
Six concrete moves that build AI-proof assignments without deploying biased surveillance:
Run the two-bar test on every assignment
Ask: Can a student cheat on this? (If yes, redesign using the “can't” principles.) And: Would a student want to? (If no, redesign using the “won't want to” principles.) Both bars must clear.
Replace at least one take-home essay per unit with an in-class equivalent
Move the high-stakes cognitive work into the room. The essay becomes a draft; the oral defense or seminar is the assessment.
Add a process artifact to every major submission
Draft history, annotated bibliography, or revision reflection. The process becomes evidence of learning — and the thing AI can't fabricate.
Name the cognitive work and the transfer for students
Tell students what thinking this builds and where they'll use it again. “This builds evidence evaluation — the skill that keeps you from being fooled.” Understanding the point is the first defense against outsourcing.
Set an explicit AI-use tier for every assignment
Tell students whether AI is off-limits, an editor, a research assistant, or the subject of the work. Ambiguity is where cheating hides; clarity is where integrity lives.
Have the conversation — don't just post the policy
Discuss cognitive offloading with students: when AI augments their thinking and when it offloads the thinking they need to do. A student who understands the difference won't want to cheat.
Section 8: The Tool That Runs the Test
Run the Two-Bar Test Before You Assign, Not After
The honest case for a product here is that it makes the ethical path the easy path. Secondary AI doesn’t build a better detector — it builds the design tool that prevents the need for one.
AI-proof assignment scoring
Secondary AI analyzes an assignment and scores how resistant it is to AI completion — then suggests concrete revisions using the principles above. Run the two-bar test before you assign, not after.
Socratic chatbots that build the skills students won't want to outsource
A Socratic chatbot guides a student through productive struggle — building the very skill the assignment is designed to develop. A student who built the skill wants to use it; a student who didn't is tempted to fake it.
Process-based assessment tools
Exit tickets, discussion prompts, and reflection tools that make student thinking visible — so you assess the work happening, not the artifact submitted. The process artifact, built in.
Differentiation that levels up, not down
Scaffolded, on-level, and extension versions of the same rigorous task — so every student clears the same cognitive bar through the right path. No one gets the “easier version” that teaches them their thinking doesn't matter.
The honest caveat: The tool scores and suggests; the judgment is yours. A high AI-resistance score doesn’t make an assignment good — it makes it uncheatable. You still have to make it worth doing. The two-bar test is a design discipline, not an algorithm. The tool helps you run it; it doesn’t run it for you.
Section 9: Frequently Asked Questions
The Design Question, Answered
Design tasks that structurally require what AI can't produce: oral defenses (AI can't speak for you in the room), visible process (draft history, reflection), local context (your school, your data, this week), in-class collaboration, and staged iterative submissions. The cheating path is closed by design, not by threat of detection.
Continue Reading
Design Assignments Worth Doing — With AI Built In
Run the two-bar test on every assignment, set an explicit AI-use tier, and deploy Socratic chatbots that build the skills students won’t want to outsource.