Assessing Student Work with GenAI Tools: Considerations for Instructors

Assessment includes a broad range of practices and instruments, but a fundamental distinction important in this context is the difference between grading and feedback. These two activities serve different purposes and affect student learning and educational trajectories in different ways.

Grading

Grading refers specifically to evaluating an assignment, activity, or assessment by awarding points, scores, or rankings that will be factored into a student's final course grade. Final course grades affect students' GPAs, which can impact their scholarship or grant eligibility, future admissions to educational programs, and their standing within their programs. 
UMN Policy on Grade Accountability for for-credit courses established that grading is the responsibility of the primary instructor. 

“The instructor who is in overall charge of a course offering is accountable for all grades given to students. Responsibility for grading or evaluating student work in a course may be assigned to a teaching assistant or grader, but ultimate responsibility remains with the course instructor. All individuals who grade or evaluate student work in a course must have a formal affiliation with the course (e.g., as instructor of record, teaching assistant, paid grader).” 

The specificity of UMN policy and the high-stakes nature of grading in university courses suggest that instructors should proceed very cautiously with using AI to assign course grades or rank students’ overall performance. While electronic tools can be used to assess multiple-choice or true/false exams or single-answer objective questions, the ability to read, understand, interpret, and assess students' responses to more complex questions may still be many years away. Similarly, while GenAI tools can distinguish correct and incorrect answers, they are less capable of understanding why a student might have answered incorrectly, and are unlikely to identify the conceptual or procedural mistake that caused a student to go awry.

Feedback

Feedback includes activities where a student receives coaching and instruction based on a review of their performance. Instructors can offer feedback during the work-in-progress stage (formative feedback, intended to encourage revision) or at the end of the process (summative feedback, intended to offer future direction). Feedback can include comments on student work (Oral or written, in person or asynchronous, live or recorded), preliminary rubric scores, practice tests, and coaching after a skill demonstration. 

Typically, the role of feedback is to help students refine their understanding or skill development and progress more successfully through a course. For that reason, more opportunities for feedback can improve student learning and performance (whether from the instructors, peers, or trained GenAI respondents).

GenAI tools can supplement instructor feedback. In contexts with limited or clearly defined outputs, such as producing functional code or writing a structured response to a short-answer question, an instructor might use an AI tool to check a batch of coding assignments for syntax errors to give students rapid formative feedback, but the instructor makes sure to manually review the logic and, ultimately, awards the final grade themselves.

Similarly, because GenAI tools are powered by large language models, they can provide guidance on grammar, spelling, and usage, as well as many other conventional language choices. The effectiveness of AI in assessing higher-order thinking skills, disciplinary, procedural, and conceptual knowledge, and other complex learning outcomes remains uncertain, although researchers in the scholarship of teaching and learning are innovating widely and now publishing their results.

Decisions about whether and how to incorporate GenAI feedback into your courses should align with your teaching philosophy, course context, and student learning goals. The following questions can help you think through your approach, identify key decision points, and be prepared to explain to students the role of GenAI feedback in your course.

• Why am I interested in using GenAI tools?
  • Score objective assessments, like homework problems or multiple-choice questions, quickly and efficiently?
  • Reduce the workload for instructors, teaching assistants, or other reviewers?
  • Eliminate discrepancies across assessments, assignments, or sections of a course?
  • Provide feedback more promptly than I am otherwise able? 
  • Give students formative feedback or self-guided opportunities for practice? 
  • Meet another need within my specific context?

Identifying the pedagogical rationale for GenAI integration is a critical first step before adopting a new technology. Instructors should develop and communicate a comprehensive plan to their students that aligns the use of GenAI with learning objectives, ensures transparency, and maintains rigorous oversight. 

Consider the following strategic questions as you develop a plan for using GenAI for feedback:

• How does GenAI fit within my current approach to assessment? How does providing feedback help me stay informed about what students are learning and how they are progressing through the course?
• Could other tools, technologies, or platforms automate feedback without AI (e.g., creating an auto-graded Canvas Quiz)? 
• How will a GenAI tool augment human feedback, helping students meet learning goals, and encourage growth and improvement?
• What forms of response can generative AI offer  (e.g., in-text comments, corrections, overt revisions of student text, endnote comments, rubric estimates, etc.)?
• How can I ensure that AI-generated feedback remains fair, unbiased, and aligned with individual student needs?
• How will I verify GenAI feedback or ratings?
• Are students allowed to opt out of GenAI feedback on their work (for ethical, environmental, or personal reasons)? How will you ensure students who opt out will receive equivalent opportunities for feedback?
• How will I support students in interpreting or acting on GenAI feedback? What will the process be if/when students question grades or credit earned for AI-related feedback activities?
• How will I communicate the role of GenAI in the course to my students and model the practices of AI disclosure I expect?
• Is my own use of GenAI consistent with my course policies on GenAI? How might the use of GenAI for feedback impact students' motivation, sense of trust, and instructor credibility?
   

The University of Minnesota encourages instructors to maintain instructional integrity and foster student-instructor trust. As such, GenAI should function as a supplement to instructor judgment, not a replacement. To ensure equity and academic integrity, it is crucial that feedback and assessment remain centered in human discernment and oversight. 

GenAI tools are inherently limited in their ability to contextualize student learning progress. While GenAI may be (mostly) successful at identifying single correct answers for closed-ended tasks or questions, it lacks the nuance required to evaluate complex, open-ended assignments. 

To ensure instructional integrity, adhere to the following principles:

• Verify GenAI feedback for accuracy, consistency, and bias by actively participating in the feedback process. Human review of AI output is essential.
• Maintain your presence in the assessment process by comparing GenAI feedback against your own assessment of students' work to identify hallucinations (untrue statements generated by GenAI tools). GenAI may confidently but incorrectly assert factual errors or misinterpretations, (e.g. claiming that a student missed some feature of a response that is actually present.)
• Prioritize the ethical use of GenAI by practicing transparency, adhering to the University's data privacy and security standards, and promoting human engagement and involvement.
• Include activities that encourage or facilitate student agency to engage with feedback by summarizing, evaluating, and acting on it (e.g., exam wrappers, revision memos, and assessments of feedback quality and value).
   

The following section identifies several risks and ethical benchmarks for instructors seeking to integrate GenAI into their feedback workflows while upholding University standards and pedagogical integrity. Unfortunately, this list is not exhaustive (and is likely to be revised as we learn more about the benefits and risks of AI feedback)

Consider individual students' rights to their intellectual property: Because students' intellectual property is their own, be exceptionally clear with students about how and why you are incorporating GenAI tools to assist with feedback and obtain their consent to have their documents uploaded to an AI. Students have the right to refuse to have their intellectual property uploaded to any GenAI tool.

Be transparent about the complexity of effectively prompting GenAI tools to align with desired outcomes: This transparency has the added benefit of clarifying for students just how much labor is involved in designing and refining prompts that yield effective GenAI responses. Explain to students what tool(s) you are using, how you set them up to provide relevant feedback, and how you will play an active role in monitoring or quality-checking the GenAI output. Detailed and specific feedback requires that instructors provide the GenAI tool with fine-tuned instructions on the assignment's genre expectations. Using GenAI to augment feedback is not a strategy to avoid hard work; it's a labor-intensive process intended to produce better outcomes.

If employing GenAI for feedback, explicitly emphasize the role of the GenAI feedback alongside human review in supporting learning outcomes, and be clear about what students should do with the feedback they receive. 

Be wary of the unintended consequences of outsourcing: UMN focus group research indicates that students are more likely to use GenAI inappropriately in courses they consider peripheral to their interests or where they perceive ‘busywork.’ They may also be more inclined to outsource their academic work to a GenAI tool if they perceive that their instructor is doing the same. Students may also presume that resorting to GenAI for feedback is evidence that the instructor is less interested in or committed to their students. A recent study by Morris and Maes (2026) found that learners who believed a real person had given them feedback spent more time on future assignments than those who were told GenAI had given it, even though the feedback was identical and both groups described it as helpful.

Exercise diligence with third-party tools: The educational technology market is saturated with GenAI solutions whose marketing claims may exceed their performance. Tools often perform well on standard tasks but may fail when encountering the complex variations inherent in student work. Furthermore, instructors must ensure that any third-party tool complies with the University's data privacy and security requirements. The rule of caveat emptor applies in this environment: don't be surprised if a technology vendor promotes the benefits of a feedback tool without supplying clear evidence of its effectiveness in promoting student learning. Further, vendors will almost certainly ignore the drawbacks and unintended consequences that customers may incur.

Please review the UMN guidelines for the appropriate use of artificial intelligence for further guidance.
 

Instructors are strongly encouraged to model the attribution practices they expect of their students. If you are considering using GenAI feedback tools:

• Disclose your intention to use GenAI feedback in your course policies and discuss your motivations for integrating this practice. 
• Include Artificial Intelligence Disclosures when you use GenAI in your writing process.
• Allow students to opt for a human reader if they object to receiving GenAI feedback.
• Select the most robust reasoning models when using a GenAI tool for feedback. 

While it may be tempting to use prior student work to test the accuracy and validity of GenAI feedback tools, do not upload any student-generated content without the consent of the student author. Ask the student's permission before using their responses to fine-tune a customized chatbot or another AI agent.

Stress that you are augmenting feedback, not replacing it: To reassure students of the accuracy and effectiveness of the feedback they receive, explicitly emphasize that human experts will review and test the AI-generated feedback. Explain how GenAI feedback may inform the final course grade you assign students. 

As you explore using GenAI tools for feedback, consider first adding a round of GenAI feedback on an assignment in addition to your current evaluation system, rather than replacing it. For example, add a step to an existing assignment where students submit a draft of their writing to a GenAI tool and compare the GenAI feedback they receive with human feedback on the same draft. What differences emerge in accuracy, tone, or focus? Where is GenAI effective (such as copyediting or providing explanatory detail) and where is it less effective (such as understanding why a student's answer misses the mark or other elements of judgment)?