Reviewing Exam Mistakes with AI

After an exam, a score says how many points were earned but often not what to practice next. Research on learning from errors emphasizes the role of corrective feedback in turning an error into a learning opportunity under the studied conditions. Begin with the question, the student's original work, the rubric or official answer and any instructor comments. Avoid asking AI to guess what a teacher meant from a score alone. If an item is still under review or a course restricts AI use, follow those rules first. Classify the error at the first point where reasoning goes wrong. A student may know the concept but choose the wrong method, perform a valid method with an arithmetic slip, misread a unit or use an incorrect fact. These require different follow-ups. Ask AI to compare the student’s step with the official reasoning and state the evidence for its diagnosis. Then check the label yourself; a model can mistake an alternative valid method for an error or miss a hidden earlier assumption. Keep uncertainty visible when the key itself is ambiguous. Correct the work in the learner’s own hand. Explain why the incorrect step seemed plausible, why it fails, and what cue would signal the right method next time. Try a new item of the same type without looking at the solution, followed by a different-looking item that tests the same principle. A later retry helps reveal whether the repair persisted or only the original answer was memorized. Track error patterns over several assessments instead of treating one mistake as a fixed weakness. For a disputed grade, follow the instructor’s review process; an AI explanation is not authority over the rubric. Protect other students’ work and private feedback. The useful outcome is an actionable practice plan with verified examples, not a polished post hoc story that makes every wrong answer sound obvious.