Developing AI-Resilient Assessment

These strategies emphasize the reinforcement of academic integrity expectations and provide threshold strategies for mitigating academic integrity violations. By implementing strategies, such as syllabus and assessment-level statements regarding academic integrity and the use of AI, AI-use disclosure declarations or AI and academic integrity pledges, you provide learners with clear expectations for the behavior expected of them and what is permitted in terms of use of generative AI, collaboration with peers, use of materials and study aides, etc. Because there is no universal definition of cheating, the strategies enable you to clearly articulate what is or is not acceptable.

For quizzes and written assignments, strategies, such as the use of plagiarism detection tools (e.g., Turnitin) and the use of varied question types, randomized questions, display of single questions per page, and timed assessments help to mitigate some level of academic dishonesty. However, these strategies may still be limited in mitigating unsanctioned use of AI. 

Low-effort, moderate-scalability approaches emphasize strategies that make learner thinking visible. These approaches rely on learner metacognition and transfer of learning. Metacognitive processes invite learners to evaluate their approaches to the problems they are trying to solve and to assess and reflect upon their work and to feed forward their learning to future assessments or their future work. Opportunities to revise, reflective components in assessments, and contextual assessments are potential ways in which to integrate these metacognitive components. Providing opportunities for learners to revise their work provides opportunities for them to reflect, assess their performance, and to determine where they may have made mistakes and to try new alternatives. Likewise, creating opportunities for learners to apply their learning to situations in their personal contexts enables transfer of learning, applying the same concepts in a different situation.

Strategies in this area stress growth and process over the final output. Portfolio assessments or iterative drafts enable learners to demonstrate their development over time. These approaches may leverage existing assessments and incorporate reflection or other elements to help learners contextualize their learning. Depending upon the number of documents to review, the process of reviewing and grading can be time-consuming, thus impacting scalability.

These strategies invite learners to reason and defend decisions to applied problems, while still working well in large-enrollment courses. The use of application-based multiple-choice questions enables partial automatic scoring and scalability. These Questions use realistic context-specific decisions, where multiple answers could be possible, and learners need to choose the best option based on the scenario. Adding structured ambiguity (incomplete data, competing priorities, or time pressure) increases judgment demands and reduces the value of generic AI-generated responses. Requiring a brief justification for the selected response (two to four sentences) surfaces reasoning without creating a heavy grading burden, especially if you use a tight rubric or justification “anchors” (e.g., must reference one course concept and one constraint). When grading these assessment types, you need to grade for reasoning signals (use of evidence, tradeoff recognition, concept application) rather than writing polish. Allow AI to be used as a support tool, as long as learners can defend the logic concisely.

Assessment strategies in this category use case studies and open-ended problems to enable learners to demonstrate the contextual reasoning and judgment important to post-graduation. Assessments may use a similar case structure and common deliverable templates to guide learners. Guided assignment prompts, such as, “Write one recommendation,” or “Provide three tradeoffs, two risks, and a metric plan,” ensures a consistent deliverable from each learner that can be easily graded. However, this approach also lends itself to easy reproducibility of responses among learners as well as generic responses produced using GenAI. Thus, randomizing key parameters (e.g., industry, constraints, performance indicators, customer segment, budget, and stakeholder perspective) for each learner or groups of learners ensures a unique, meaningfully different context for each assessment. Varying these parameters increases rigor and reduces opportunities for learners to easily reproduce similar responses or to rely solely on GenAI. Rubrics that prioritize assumptions, quality of evidence, and their application of logic and tradeoffs can assist with assessing thought processes vs. polish or final product while still maintaining scalability. 

These strategies—including oral examination, viva voces, interactive presentations, and ethical reasoning—surface spontaneous evaluative judgment. They work well when you want to verify that learners can explain their thinking, adapt to follow-up questions, and defend tradeoffs, rather than simply submitting polished outputs. Because learners must respond live, these formats reduce the value of pre-written AI-generated text and make reasoning gaps visible quickly. Best practice for creating these assessments require them to be short and structured (e.g., 8–12 minutes per learner or team) using a standardized question bank with a few adaptive follow-ups to maintain fairness and manage time (Monsha, 2025). To keep workload manageable and increase scalability, use only a subset of learners each cycle, group vivas with individual accountability inside a team, or “defense checkpoints” after key submissions, rather than after every assignment. Ethical prompts work best when tied to a concrete scenario (e.g., conflicting stakeholder incentives and graded on the quality of reasoning—clear principles, consequences) and justified decisions. When grading these assessments, give learners a transparent rubric and a short practice run so the live format feels like a professional skill-building experience, not a “gotcha.”

High-effort, high-scalability strategies emphasize process-oriented assessment design, including scaffolded assignments, structured checkpoints, draft submissions, and scaffolded reflections that require learners to document their thinking, decision-making, and revision process. Scaffolded assessments break a complex task into smaller, structured stages (e.g., proposal, draft, revision, and final submission), providing guidance and feedback at each step to support learning progression. Structured checkpoints are planned moments within an assessment where learners submit works in progress (e.g., outlines, drafts, data analysis, design plans) to demonstrate their thinking and receive formative feedback before completing the final product. Scaffolded reflections are reflective prompts embedded throughout an assignment that guide learners to explain their reasoning, decisions, revisions, and learning process. These strategies prioritize the development and demonstration of understanding, rather than relying solely on evaluation of the final product. They reinforce academic integrity by increasing learner transparency and accountability. The staged, iterative structure supports deeper learning and skill development and makes inappropriate or undisclosed AI use more difficult. These strategies do require significant upfront instructional effort to plan, align, and design feedback mechanisms. However, once structured templates, clear rubrics, and repeatable checkpoints are established, they can be highly scalable and reliable.

These strategies emphasize the development of higher-order thinking skills with activities, such as metacognitive tasks that intentionally incorporate AI, multimodal submissions, and design-based projects. Metacognitive tasks incorporating AI are activities wherein learners use AI as part of their work and then explicitly reflect on how they used it, why they made specific choices, what they accepted/rejected, and what they learned, making their thinking and judgment visible. Multimodal submissions are assessments that require learners to demonstrate learning through more than one format (e.g., written analysis + video explanation + visuals/data/artifacts), providing multiple kinds of evidence of understanding. Lastly, design-based projects are assessments in which learners create, test, and refine a solution, product, or intervention for a defined problem, typically involving iteration, justification of decisions, and evaluation against criteria or stakeholder needs. These strategies encourage learners to analyze their decisions, justify their use (or non-use) of AI tools, and reflect on how AI shaped their reasoning and outcomes.

By emphasizing analysis, synthesis, and creation, this approach shifts the focus from task completion to cognitive engagement. At the same time, it increases transparency around AI use within the learning process, reinforcing academic integrity through applied, reflective, and process-focused work. While this approach requires substantial instructional effort in design, feedback, and evaluation, its scalability is moderate; it can be expanded more effectively when clear rubrics, structured reflection prompts, and adaptable project frameworks are used to support consistent implementation.

High-effort, low-scalability strategies emphasize learner application of human-centered design concepts by situating assessment in authentic, relational, and context-rich learning experiences. Strategies may include community-engaged or capstone projects or performance-based assessment situated in real-world contexts. Learners may apply course concepts to projects that involve local or professional contexts and community partners or real clients. These types of experiences require learners to develop contextual understanding, consider stakeholder input, and assess practical constraints. Projects may happen within the context of the course, or, as in the case of capstone projects, they may be the culmination of a learner’s knowledge and skills development across a program. Capstone projects are a culminating assessment, usually near the end of a course or program, wherein learners integrate and apply learning across multiple concepts or courses to complete a substantial project (e.g., research, design, or professional deliverable), often including reflection and/or a formal presentation.

While scalability can be challenging due to the time, coordination, and feedback required, it can become more manageable when work is structured collaboratively, with responsibilities distributed across teams rather than assigned solely to individuals.

Dr. Woods has 6 weeks to prepare her course for the next term. She is aware that learners in her college are increasingly using GenAI for to complete portions of their assessments. With the course coming up fast, Dr. Woods decides to introduce small changes for this run of the course to help address learner’s AI use. First, she adds the college’s approved syllabus statement to identify for learners the acceptable uses of AI for her course. Dr. Woods also identifies two of case study assignments for which learners answer a series of fact-based questions about the case and that could easily be generated by loading the case into a GenAI tool. She rewrites the assignments. Learners still get the basic questions, but she has added additional questions in which they need to analyze the case, make a recommendation for how they would respond if they were consulting the CEO, and to explain the rationale for their decisions. Likewise, she adjusts the marking for the assignment to assess the learner’s ability to think critically about the topic and to generate and defend their solution.

Dr. Pembroke’s ACCY 989 course is scheduled to be redesigned in the next 6 months. This is an opportunity to significantly overhaul the course assessments. Working with his learning designer, Dr. Pembroke decides to scaffold his assignments. Each assignment builds on the next. Each assignment includes a detailed rubric to provide feedback to learners and direction for what they may consider doing differently. Subsequent assignments revisit skills, providing learners opportunities to improve on their previous work. The assignments also include an opportunity for learners to describe what they did differently on the assignment than in the previous attempt to help the graders and instructional staff help to address learner misconceptions.