Designing Assessments and Learning Activities That Discourage AI Cheating
If institutions have reached one conclusion quickly, it鈥檚 this:
The most effective response to AI-related cheating is not surveillance鈥攊t鈥檚 design.
A more productive question than 鈥淗ow do I stop students from using AI?鈥� is:
鈥淗ow do I design learning activities where outsourcing the thinking simply doesn鈥檛 work?鈥�
This reframing shifts energy away from enforcement and toward instructional control. Faculty cannot fully control technology. But they can control assignment structure, visibility of learning, and clarity of expectations.
When assessments are meaningful, contextual, and process-oriented, AI becomes either irrelevant鈥攐r obviously insufficient.
Many traditional assignments evaluate a polished final product: a paper, a discussion post, a problem set.
That model worked well in an environment where producing polished work required visible effort. Generative AI changes that dynamic because it can generate structurally coherent, grammatically polished content quickly.
Generative AI is very good at producing polished products.
It is much less effective at demonstrating:
鈥� Thinking over time
鈥� Personal engagement
鈥� Iterative revision
鈥� Course-specific nuance
These elements reflect learning, not output.
When assignments focus only on the final product, AI can mimic success.
When assignments reveal process, AI becomes far less useful as a shortcut.
This isn鈥檛 about redesigning everything. It鈥檚 about making thinking visible.
You do not need a course overhaul to reduce misuse. Small structural changes often have an outsized impact.
1. Break One Major Assignment into Stages
Instead of:
Submit final paper.
Try:
鈥� Topic proposal (short paragraph)
鈥� Annotated source list
鈥� Outline
鈥� Draft
鈥� Reflection on revisions
鈥� Final submission
Each stage creates a record of development. AI can generate text, but it cannot convincingly simulate authentic growth across multiple checkpoints without sustained effort.
AI can help draft text鈥攂ut it cannot easily replicate your student鈥檚 development over time.
Even adding just one required draft checkpoint significantly reduces misuse.
2. Add a 3鈥�5 Sentence Process Reflection
Reflection strengthens integrity because it shifts evaluation toward reasoning.
At the end of an assignment, include:
In 3鈥�5 sentences, explain your approach to this assignment. What decisions did you make? What challenges did you encounter? What feedback influenced your revisions?
This simple addition:
鈥� Reveals understanding
鈥� Creates documentation of thinking
鈥� Encourages metacognition
鈥� Deters fully outsourced work
Reflection is not 鈥渇luff.鈥� It鈥檚 visibility.
It also aligns with best practices in learning science: students deepen understanding when they articulate their reasoning.
3. Make Prompts Specific to Your Course
AI performs best when prompts are broad and generic. It performs less convincingly when assignments depend on course-specific interaction.
AI struggles with:
鈥� Course-specific case studies
鈥� Your lecture examples
鈥� Class discussions
鈥� Ongoing semester themes
Instead of:
Explain Theory X.
Try:
Apply Theory X to the case study discussed in Module 3 and connect it to the debate example from Week 5.
Specificity reduces generic AI output.
It also strengthens alignment between assessment and instruction鈥攊mproving learning while reducing misuse.
4. Shift Weight from One High-Stakes Task to Several Smaller Ones
Assessment structure influences behavior. When one assignment carries extreme weight, anxiety rises鈥攁nd anxiety increases the likelihood of shortcuts.
High-stakes assignments create pressure.
Pressure increases misuse.
Instead of:
鈥� One 40% paper
Consider:
鈥� Proposal (5%)
鈥� Draft (10%)
鈥� Peer feedback (5%)
鈥� Final submission (20%)
Lower pressure = lower temptation.
Online environments are especially well-suited to distributed grading because learning management systems make staged submissions easy to manage.
5. Allow AI in Structured Ways (Instead of Pretending It Doesn鈥檛 Exist)
Prohibition without clarity often drives hidden use. Structured inclusion can be more effective than silence.
Ambiguity creates loopholes.
Some instructors intentionally build AI into learning:
Example:
Use an AI tool to generate an initial answer. Then critique it. Identify inaccuracies, missing nuance, or bias. Revise the response and explain your improvements.
Now AI becomes an object of analysis鈥攏ot a shortcut.
When use is explicit and structured, misuse decreases elsewhere because students understand boundaries.
6. Add Brief Oral or Multimodal Components
Authenticity increases when students must articulate their reasoning in their own voice.
Even small additions increase integrity confidence:
鈥� 2-minute recorded explanation of reasoning
鈥� Screen-recorded walkthrough of a solution
鈥� Audio reflection
鈥� Annotated screenshot explanation
These don鈥檛 need to replace written work. They simply supplement it.
Even one oral checkpoint in a semester dramatically increases authenticity.
They also improve engagement and give students alternative ways to demonstrate understanding.
7. Replace 鈥淧olish鈥� with 鈥淧rogress鈥� as a Grading Value
Rubrics send powerful signals. When grading criteria emphasize formatting and surface quality, AI can excel. When criteria emphasize reasoning and growth, AI loses its advantage.
AI excels at polish.
Human learning shows growth.
Consider grading language such as:
鈥� Evidence of revision
鈥� Depth of analysis
鈥� Engagement with feedback
鈥� Original connections
When the rubric rewards thinking, not just formatting, AI loses its advantage.
This subtle shift aligns grading with learning rather than presentation.
No assessment is cheating-proof. Technology will continue to evolve.
The goal is not elimination. It鈥檚 reduction.
Cheating-resistant design increases friction for misuse by:
鈥� Making thinking visible
鈥� Increasing relevance
鈥� Lowering panic pressure
鈥� Clarifying expectations
鈥� Valuing process
When students feel connected and supported, most choose engagement over shortcuts.
Design is not about mistrust. It is about instructional strength.
Faculty do not need to redesign entire courses to see improvement.
Start with one of these:
鈥� Add a reflection question.
鈥� Add one draft checkpoint.
鈥� Rewrite one prompt to be more specific.
鈥� Add one multimodal explanation.
鈥� Adjust one rubric category to emphasize reasoning.
Iterate next semester.
Design evolves.
Sustainable improvement is incremental.
Before the next major assignment, ask:
鈥� Does this assignment reveal thinking or just product?
鈥� Would AI-generated work be obvious in this design?
鈥� Is this high-stakes pressure unnecessary?
鈥� Is the prompt specific to my course?
If you answer yes to even one improvement opportunity, you have a starting point.
You do not need perfection. You need intentionality.
For concrete examples and models:
- University of Michigan 鈥�
- Cornell University 鈥� :
- University of Minnesota 鈥�
- EDUCAUSE 鈥� Rethinking Assessment with AI:
These institutions consistently emphasize assessment redesign as the primary instructional response to generative AI.
The most effective response to AI-related cheating is pedagogical鈥攏ot technological.
When assessments prioritize reasoning, reflection, and development, AI becomes either unnecessary or clearly insufficient.
Thoughtful design doesn鈥檛 just protect integrity.
It improves learning.