From Policing to Possibility: What GenAI taught me about Assessment
Learn how one educator has shifted from trying to police AI use to redesigning assessments that emphasize meaningful learning and student engagement. By creating authentic, scaffolded, and relevant assessments students are more likely to choose genuine participation over unpermitted AI use.
By Jan Gelech, USask Assessment ChampionWhen generative AI first became widely available, I reacted much like many instructors did: I saw a threat to academic integrity and wondered how I could prevent students from using it when they were supposed to complete work independently. Before long, I found myself becoming someone I did not particularly want to be as an educator—agitated, suspicious, and increasingly focused on policing.
Could I design assignments that AI could not complete?Could I identify AI-generated work?
Should students write everything by hand?
For my circumstances, this approach was frustrating and ultimately unproductive. AI-detection tools were unreliable. Students could circumvent many restrictions. Strategies that increased surveillance also risked damaging trust, creating accessibility barriers, and consuming enormous amounts of instructor time. Most importantly, a singular focus on policing distracted me from the central purpose of assessment: supporting learning.
Eventually, I began asking a different question. Rather than asking, “How can I control students’ use of AI?” I asked, “How can I create assessments that students consider worth doing themselves in a world where AI exists?” To explore this question, I reflected on patterns across my own courses.
Unpermitted AI use1 appeared more likely when an assessment was high stakes, impersonal, easy to outsource, or focused almost entirely on the final product. It appeared less common when students found the work personally or professionally meaningful, interacted with me throughout the process, received support and feedback, and needed to apply ideas to specific or novel contexts.
For example, students in my qualitative research course complete highly scaffolded projects based on questions they help develop. They work through interconnected stages, make methodological decisions, receive formative feedback, and explain their reasoning. The value of the assignment lies not only in the final report but also in doing the intellectual work along the way. In contrast, a generic assignment completed privately and submitted only once at the end offers far more opportunity—and sometimes far more temptation—to offload the work.
As I continue to confront the challenges of generative AI, I have begun thinking of assessment redesign more in terms of “tuning” than policing or wholesale replacing. Borrowing from the world of sound engineering, I imagine we can adjust numerous dials and levels to change our (behavioural) outputs: stakes, visibility, relational accountability, instructional scaffolding, formative feedback, student confidence, personal and professional relevance, ease of offloading, and emphasis on process rather than outcome.
A wealth of research evidence suggests that turning even one or two of these dials can change how readily students turn to unpermitted generative AI use. We might distribute grades across several smaller assessments, require students to document important decisions, connect tasks to local or personally meaningful issues, or build in a brief conversation about submitted work. We can scaffold complex assignments, provide feedback before the final submission, allow revision, or ask students to reflect on how their thinking changed. Authentic tasks designed for a particular audience with known goals and aspirations can also make completing the work more valuable than merely producing an acceptable product.
Of course, none of this creates an “AI-proof” assessment, and not every dial can—or should—be adjusted in all circumstances. Large classes, limited resources, learning objectives, accessibility needs, and instructor workload all constrain what is possible. Changes can also have unintended consequences. A redesign that discourages AI use but makes an assignment less accessible for certain students or unmanageable for the instructor is not an improvement.
Still, this shift in thinking has restored some of my sense of agency. I cannot control every tool students use or every decision they make. I can, however, create conditions in which students are more likely to feel that they can do the work, that the work matters, and that their thinking will be seen. Perhaps, for me, effective assessment in the age of generative AI is not primarily about making unpermitted use impossible. Perhaps it is about doing what I can to make authentic engagement the most reasonable, worthwhile, and appealing choice.
1 When is unpermitted AI use MORE likely? Gelech, J. (April 30, 2026). AI as a catalyst for better assessment design. USask Assessment Conference, Saskatoon, SK, Canada.
About the USask Assessment Champions
This article is part of a series featuring USask Assessment Champions, educators recognized for assessment practices that help students learn deeply, demonstrate their abilities, and thrive academically. Discover more about the Champions and their work on the Assessment Champions webpage.
Dr. Jan Gelech is a USask Assessment Champion and a Lecturer, in Psychology and Health Studies at the College of Arts and Science.
Title image credit: stevepb on Pixabay.com