Roundup on Research: Building Trust into Your Assessments in the Age of AI

Richard Marks is a curriculum and assessment specialist with the Center for Research and Learning and Teaching in Engineering.

Generative AI is becoming a common resource for many students, but their ability to use it productively and safely can vary. Overall, students tend to have a positive view of GenAI and are enthusiastic when it is integrated into their learning experiences (Heil et al., 2025). However, we have also been hearing more complex views in a range of settings at U-M. In focus groups and disciplinary discussions, many students express concerns about issues such as environmental impact, bias, and academic integrity.

It is also important to distinguish between enthusiasm and understanding. While students may have mixed feelings about using GenAI in their learning, they may not always have a full understanding of the technical, legal, and learning implications of GenAI tools. Conversely, some research suggests that students with the highest levels of GenAI competency find it to be the most supportive of their learning experiences (Heil et al., 2025).

This highlights the need for instructors to support students’ GenAI competency with the goal of ensuring that all students can benefit equitably from using it. By helping students build their GenAI literacy skills, instructors can ensure that more students benefit and fewer students are exposed to avoidable risks. Doing so, however, depends on creating a climate of trust. That means creating an environment in which students feel empowered to explore and better understand tools that can impact their learning without the fear of being surveilled or punished for doing so.

The Importance of Trust

Assessment hinges on trust. When instructors cultivate a climate of trust, students feel greater freedom to express and explore their ideas, which can lead to more authentic engagement with an assessment. Building trust with students is an essential component of effective instruction, as trust contributes to the meaningful interactions in the classroom, which supports learning (Pederson et al., 2022). In fact, students have reported that feelings of trust and psychological safety were among the most important factors helping them navigate challenging STEM environments in higher education (Pederson et al., 2022). 

The trust that students feel in the classroom can be negatively impacted by an atmosphere of fear associated with GenAI use (Luo, 2025), and this distrust can undermine the integrity of assessments (Carless, 2009). While concerns about GenAI promoting academic misconduct are common, evidence suggests that college students have a preference for GenAI tools that enhance their understanding rather than simply generating or paraphrasing content (Heil et al., 2025). Cheating is not a new phenomenon in the field of education. Furthermore, instructors cannot rely on tools for AI detection. Current AI detection tools lack accuracy, and can be easily manipulated or bypassed (Perkins et al., 2024). Their use can also erode the trust between students and instructors. 

Altogether, the emerging literature base is painting a picture of college students who seek out GenAI tools that will support, not replace, their learning. What they lack is a strong, theoretical understanding of GenAI and would benefit from trusted instructors who provide a safe space to explore it. In practice, trust is sustained when students can clearly anticipate what is expected, and instructors are not forced into an enforcement role they cannot reliably perform. Because AI detection has limited effectiveness, the most trust-preserving approach is transparent course communication paired with assessment designs that promote authentic engagement with the learning goals. What follows are policies and design choices that clarify expectations for students and create assessment conditions that support authentic work

Establishing a Clear GenAI Policy

Building trust around GenAI in higher education begins with establishing a transparent and comprehensive GenAI policy for your course. The lack of a clear policy and two-way communication between students and instructors compounds challenges in higher education assessment (Luo, 2024). Without this communication, students may lack understanding of when and how GenAI can be used, whether it must be disclosed to the instructor, and if they will face negative consequences for using GenAI on a given assignment. Currently, many policies in higher education are surveillance focused, policing students’ GenAI use (Luo, 2024). Instead, policies should promote transparency to avoid framing students as inherently dishonest.

When developing a GenAI policy, instructors should consider what constitutes “originality” in their course. While a traditional view of originality places emphasis on solely self-generated work, GenAI is increasingly integrated into tools such as Microsoft Office and becoming more intertwined with productivity. Therefore, when deciding what makes a student’s work “original,” instructors should first consider the disciplinary context and learning goals to identify which aspects of student work must be solely self-generated to demonstrate progress toward learning outcomes (Luo, 2024).

GenAI Policy Example

Common Frameworks for Assessments

When designing assessments, it is important to realize that they cannot be “AI-proofed.” Instead, instructors can design assessments that are less vulnerable to GenAI being used in ways that undermine the intended learning goals, or that intentionally incorporate GenAI in ways that support learning. The “traffic light” system uses red, amber, and green to indicate permitted levels of GenAI use across different assessments. Declarative frameworks are another common approach, which ask students to document and disclose their use of GenAI in assessments (Corbin et al., 2025). The limitation of these common frameworks is the reliance on student compliance. To reduce this reliance on compliance, Corbin et al. (2025) outline two categories of assessment changes aimed at maintaining validity in GenAI contexts: discursive changes and structural changes.

Discursive changes are those which alter the way in which students are instructed to approach or complete a task, without changing the actual assessment. Discursive changes may involve alterations to the communication of instructions, rules, or guidelines to students, and the success of these changes depends on students being aware of and understanding them. 

Structural changes, however, directly alter the nature, format, or mechanics of a task. By making structural changes to assessments, instructors can build validity into the design itself rather than relying solely on language and voluntary compliance.

It is important to note that communication is required when any changes are made to assessments to ensure that the instructions and structure of the assessment are clear to students. While discursive changes communicate the rules and expectations governing students’ approach to an assignment, structural changes communicate that the actual nature of an assessment has been altered. Both require students to be provided with instructions, but structural changes ensure that the assessment’s validity does not depend entirely on students’ willingness to comply with them (Corbin et al., 2025).

Assessment-Specific Strategies

In practice, instructors can incorporate these structural and discursive changes into a variety of assessments, clearly directing learners to complete them with the proper implementation of GenAI while also demonstrating metacognitive thinking and knowledge acquisition.

Written assessments

• Clearly communicate guidelines surrounding GenAI usage
  • What is allowed, in addition to what is not allowed
  • Transparency and disclosure requirements
  • Example prompts
• Submit in stages (e.g. rough drafts, annotated bibliographies), and teach aspects of the writing process with each stage
• Design collaborative activities
  • Peer review and response: Critique or build on peers’ contributions 
  • Group grant/proposal writing, mirroring common professional collaborations
• Incorporate reflections on the learning process

Online discussions and assessments

• Include reflective questions: Prompt students to connect material to personal experiences or current events
• Socratic questioning: Why do students hold a certain viewpoint? How can conflicting evidence be reconciled? Strengths and weaknesses of an argument?
• Research-based discussions: Source recent studies or articles not likely to be found in AI searches

Conclusion

Ultimately, assessment in the age of GenAI depends more on designing meaningful ways for students to demonstrate their learning and less on monitoring for academic misconduct. The development of clear GenAI policies and two-way communication can create an atmosphere of trust, while making structural changes to assessments can protect validity without chasing a reliable way to police GenAI use. By helping students understand what is allowed, why it matters, and how an assessment is designed to demonstrate learning, students’ use of GenAI becomes a learning opportunity instead of a compliance problem.

Richard Marks is a contributor to the Center for Research on Learning and Teaching’s Substack, CRLT @ UMich

Resources

Redesigning Assignments

Discussions and Strategy

References

Carless, D. 2009. Trust, distrust and their impact on assessment reform. Assessment & Evaluation in Higher Education 34(1), 79–89.

Corbin, T., Dawson, P., & Liu, D. (2025). Talk is cheap: why structural assessment changes are needed for a time of GenAI. Assessment & Evaluation in Higher Education, 50(7), 1087-1097.

Heil, J., Ifenthaler, D., Cooper, M., Mascia, D.L., Conti, R., & Penna, M.P. (2025). Students’ perceived impact of GenAI tools on learning and assessment in higher education: the role of individual AI competence, Smart Learning Environments, 12, Article 37.

Luo, J. (2024). A critical review of GenAI policies in higher education assessment: a call to reconsider “originality” of students’ work, Assessment & Evaluation in Higher Education, 5(49), 651-664.

Luo, J. (2025). How does GenAI affect trust in teacher-student relationships? Insights from students’ assessment experiences. Teaching in Higher Education, 30(4), 991-1006.

Pederson, D.E., Kubatova, A., & Simmons, R.B. (2022). Authenticity and psychological safety: Building and encouraging talent among underrepresented students in STEM. Teaching & Learning Inquiry, 10

Perkins, M., Roe, J., Vu, B.H., Postma, D., Hickerson, D., McGaughran, J. & Khuat, H.Q. (2024). Simple techniques to bypass GenAI test detectors: implications for inclusive education. International Journal of Educational Technology in Higher Education, 21, Article 53.