AI-Assisted Grading, Part 2: What Happened When I Tried It
Earlier this year, I wrote about an experiment I planned to run in one of my courses: using AI to support my grading process. The premise was straightforward. I would remain fully responsible for evaluating student work, but I would use AI as a tool to help organize, refine, and expand my written feedback.
The goal was not efficiency alone. It was to improve the quality, clarity, and usefulness of the feedback students receive.
Now that the semester has concluded, I can evaluate how that experiment actually performed.
What I set out to do
In the original article, I outlined a few specific goals:
Maintain full human control over grading decisions
Use AI to structure and refine feedback, not generate judgments
Increase the depth and clarity of feedback to students
Reduce redundancy and improve organization
Ideally, save time without sacrificing quality
At its core, this was not about replacing any part of the teaching process. It was about improving one of the most time-intensive and often inconsistent parts of teaching: written feedback.
What actually happened
The results were stronger than I anticipated.
First, the time savings were substantial. Over the course of the semester, I estimate that I reduced grading time by approximately 50 percent. That alone would be meaningful, but the more important outcome was what happened to the feedback itself.
The quality of feedback improved significantly.
Using AI as a structuring tool allowed me to take rough, sometimes fragmented thoughts and turn them into organized, coherent, and professional responses. Instead of brief comments or repetitive notes, students received detailed explanations of both strengths and areas for improvement.
In many cases, feedback expanded to one to two pages per assignment. Despite taking less time to produce, it was far more comprehensive.
Equally important, the feedback became more precise. I was less repetitive and more intentional in how I communicated key points. The result was feedback that was both tighter in structure and more expansive in substance.
Impact on student learning
The most important question was whether this actually helped students learn.
From what I observed, the answer is yes.
Students were given more actionable and specific guidance. Rather than general comments, they received clear explanations of what was working, what was not, and what they could do next. This appeared to translate into stronger revisions and improved overall performance across the semester.
One shift that stood out showed up during office hours.
Students came in better prepared.
Instead of broad or vague concerns, conversations became more focused. Students referenced specific feedback points, asked targeted questions, and engaged in more productive discussions about their work. The feedback did not replace those conversations, but it made them more effective.
Student response (initial observations)
Throughout the semester, I received informal feedback from students through Canvas comments and office hour conversations.
The consistent theme was appreciation for the level of detail.
Students noted that the feedback was thorough, specific, and helpful in guiding their revisions. Several expressed that it gave them a clearer understanding of expectations and how to improve their work.
In short, the feedback was not just longer. It was more useful.
I will provide a more detailed breakdown of formal student feedback, including themes and direct quotes from course evaluations, in Part 3.
What I learned
This experiment reinforced a few important points.
First, AI is most effective when used to enhance human judgment, not replace it. The evaluation, decision-making, and accountability remained mine.
Second, better feedback does not necessarily require more time. It requires better structure. AI helped bridge that gap by turning unstructured thoughts into clear, organized communication.
Third, when students receive more specific and actionable feedback, they engage more deeply with their work. That alone justifies continued use.
What comes next
Based on these results, I plan to continue using and refining this approach.
There is still room for improvement. I want to further streamline the workflow, ensure consistency across assignments, and continue evaluating how students respond to different types of feedback.
In Part 3, I will share a more complete analysis of student feedback from course evaluations, including specific examples of how students experienced this approach.
For now, the conclusion is straightforward.
Using AI to support grading did not reduce the quality of my teaching. It improved it.