Academic integrity has always been a concern when facilitating online exams, but the widespread availability and rapid advancement of generative AI tools in recent years has exacerbated such concerns. In an effort to uphold the academic integrity of online exams in the age of AI, many professors have sought out online proctoring solutions. However, as outlined in Is Online Proctoring the Best Solution for My Course?, this surveillance-based software is not without flaws. The good news is that online proctoring is not the only way to safeguard academic integrity. This blog post presents some alternative assessment design strategies to consider.

Lower the Stakes of Any Single Assessment

In a significant number of cases, academic dishonesty can be driven by the stakes of a single test. A single exam that is worth 50% of the final grade creates enormous pressure for students. Research suggests that the pressure to receive a good grade on a high-stakes exam is a driver of academic dishonesty (French et al., 2024). Between the rising cost of tuition and necessity of a college degree to enter certain fields of employment, many students literally cannot afford to fail such a high-stakes exam. Spreading assessment across smaller, more frequent, lower-stakes assignments can help reduce that pressure and provide instructors with a richer picture of student learning over time.

Redesign Assessments for Application, Not Recall

Many cheating concerns exist because the assessment is testing something easily looked up, such as a definition, formula, or date. If the answer is on Google, surveillance is the only thing standing between the student and an “open internet” exam. To mitigate this, shift questions towards application, analysis, or judgment. These skills fall under the Applying, Analyzing, Evaluating, or Creating levels of Bloom’s Taxonomy.

For example, instead of asking students to “define marginal cost,” ask them to analyze a company’s pricing decision using a real (or invented) dataset. Instead of asking them to “list the stages of mitosis,” ask students to diagnose what went wrong in a flawed lab procedure. These types of questions require reasoning through a specific scenario, which can be harder to outsource, rather than simply recalling a fact.

Check out Creating Assessments from the University of Florida for some additional ideas for creating assessment questions that align with higher levels of Bloom’s Taxonomy.

Prioritize Authentic Assessments

Traditional assessments that reward memorization and regurgitation are increasingly easy to outsource. Rather than administer a single high-stakes exam that tests memorization skills, ask your students to complete a project that mirrors real-world work (Chan & Colloton, 2024). Some examples include a case study analysis, policy memo, design prototype, or data analysis report.

Authentic assessments allow students to practice real-world applications of concepts and prioritize depth over surface-level recall (Chan & Colloton, 2024). They also allow for diverse skill assessment, as such projects can be designed to naturally draw on research, collaboration, critical thinking, and communication (Chan & Colloton, 2024). This provides a more holistic picture of students’ abilities than a single multiple-choice exam ever could.

Formats that require human interaction (e.g. oral exams, presentations, interviews) are a little more difficult to hand off to an AI assistant and test soft skills that a single exam can’t, like communication, confidence, and the ability to think on your feet (Chan & Colloton, 2024).

Assess the Process, Not Just the Product

Even the most authentic, application-based assignment can be quietly outsourced to AI if all you ever see is the finished product. Process-based assessments close that gap by making the path to the answer part of what gets graded.

Some strategies for building this in:

• Drafts and checkpoints: Require students to submit an outline, a rough draft, and a revision, each with its own due date rather than within a single file with one deadline.
• Annotated work: Ask students to show their reasoning. Some examples of what this could look like include: marginal notes on a proof, comments explaining code decisions, or a research log documenting sources and dead ends. These artifacts are difficult to fabricate convincingly because they require the same understanding as the final answer, applied at every intermediate step.
• Reflection and process memos: A short, written reflection (e.g. What challenges did you encounter? What changed between the first draft and final paper? What would you do differently next time?) asks students to think about their own thinking. This is a skill AI can imitate but not verify from the outside, since only the student has access to their actual process.
• In-class or scaffolded work time: Building part of the assignment into supervised class time (even just the early drafting stage) anchors the rest of the process in verified, in-person work.

Ultimately, process-based assessment reframes the question from “is this answer correct?” to “did this student actually do the thinking?” which is a harder question for AI to help a student fake.

Prioritize Feedback Over Grades

This final recommendation pairs well with the previous suggestion of assessing the process over the product. A single letter grade or percentage collapses a whole learning journey into one number. It tells a student how much they got right, but very little about whatthey understood, partially grasped, or missed.

Research on feedback consistently finds that specific, actionable information about performance does more for learning than a score does. Hattie and Timperley (2007) describe feedback as one of the most powerful influences on learning and achievement because it closes the gap between where a student is and where they’re trying to get to. Feedback-centered approaches also tend to reinforce a growth mindset. Scholars have found that when instructors emphasize improvement and effort over a fixed measure of ability, students are more likely to treat mistakes as something to learn from rather than evidence of a ceiling on their ability (Yeager et al., 2019).

Prioritizing feedback does not require abandoning grades entirely. Treating feedback as the primary tool for learning, and grades as a secondary summary of it, can help students understand why they earned a particular outcome rather than just whatthe outcome was.

Conclusion

While there is no “silver bullet” for preventing academic dishonesty, the strategies outlined above can help! Lowering the stakes of a single assessment, redesigning questions around application, building in authentic and process-based work, and prioritizing feedback over a single score all work together to shift assessment away from a single, easily-gamed checkpoint. The goal is not to out-surveil AI, but to instead design assessments where surveillance is not the catalyst to keeping students honest in the first place.

References

Chan, C. K. Y., & Colloton, T. (2024). Generative AI in higher education: The ChatGPT effect. Routledge. https://doi.org/10.4324/9781003459026-4

French, S., Dickerson, A. & Mulder, R.A. A review of the benefits and drawbacks of high-stakes final examinations in higher education. High Educ 88, 893–918 (2024). https://doi.org/10.1007/s10734-023-01148-z

Hattie, J., & Timperley, H. (2007). The power of feedback. Review of Educational Research, 77(1), 81–112. https://doi.org/10.3102/003465430298487

Yeager, D. S., Hanselman, P., Walton, G. M., Murray, J. S., Crosnoe, R., Muller, C., Tipton, E., Schneider, B., Hulleman, C. S., Hinojosa, C. P., Paunesku, D., Romero, C., Flint, K., Roberts, A., Trott, J., Iachan, R., Buontempo, J., Yang, S. M., Carvalho, C. M., & Dweck, C. S. (2019). A national experiment reveals where a growth mindset improves achievement. Nature, 573, 364–369. https://doi.org/10.1038/s41586-019-1466-y