AI-Assisted Grading and Personalized Feedback in Large Political Science Classes — A Multi-University Randomized Trial from Houston

Grading short-answer questions at scale is one of the most time-consuming parts of teaching large university courses, which pushes many instructors toward multiple-choice tests that give students less room to develop and demonstrate critical thinking. A team from the University of Houston, University of South Carolina, Academia Sinica, the University of Michigan and Kangwon National University tested whether GPT-4 could close that gap without sacrificing grading quality.

Across four large political science courses (271 students, roughly 3,080 short-answer responses, 88% of them AI-graded), instructors built prompt templates containing their rubric, worked examples with feedback, and each new student response. The team then benchmarked the AI-assisted grades against instructor grading on measures including how well scores tracked a student's performance elsewhere in the course, how often students requested regrades (a proxy for grading errors), and how helpful students found the feedback.

The randomized trial found AI-assisted grading approximated what an instructor grading a small class would produce, without an increase in regrade requests, while freeing instructor time. The authors are explicit that this is a productivity and consistency tool for feedback delivery, not a replacement for instructor judgment - human review remained part of every grading pipeline tested.

Read the full analysis: https://sanghoon-park.com/publications/articles/heinrich-et-al-2025/