Experimenting With AI to Improve Grading Feedback in My Online Courses
Over the past decade, I have hired and managed hundreds of adjunct faculty members. One of the most persistent challenges I see, especially in online courses, is timely and thorough grading feedback.
Most instructors care deeply about their students. The issue is rarely commitment. The issue is capacity.
In online learning, I like to say that student learning happens in two primary ways.
First, students engage with instructional materials. They read articles, watch lectures, review slides, listen to podcasts. These resources are intentionally curated and aligned with the learning objectives of a given module.
Second, students complete assessments. They write papers, build projects, conduct analyses, or apply concepts in practical formats.
The first step exposes students to information. The second step reveals whether learning actually occurred.
Reading does not guarantee understanding. Watching a lecture does not guarantee mastery. That is why assessment exists. And that is why feedback is essential.
Where Grading Often Falls Short
Too often, I see grading reduced to numbers.
A few clicks in a rubric. A numerical score. A brief comment such as “Good job” or “Well done.”
That approach may complete the grading task. It does not complete the teaching task.
Students invest time, money, and energy into their education. When they submit work, they are not just asking for a score. They are asking for evaluation, reinforcement, and correction.
Effective feedback should do two things:
First, reinforce what was done well. Students are not always confident in their answers. When they demonstrate strong understanding, we should name it clearly. This builds confidence and strengthens correct thinking.
Second, correct misunderstandings with clarity. Simply deducting points is not instruction. If a concept is misunderstood, we must explain what was wrong and why. Ideally, we also point students toward specific resources or strategies to help them improve.
That is teaching.
My Traditional Grading Process
When I teach, most assignments are papers. My typical process has been:
• Annotate the document with comments throughout
• Complete a grading rubric
• Write a comprehensive summary in the overall feedback section
Students regularly comment in evaluations that my feedback is thorough and helpful. I value that feedback because I put substantial time and energy into it.
But there is a cost.
This process is slow. It is cognitively demanding. It is exhausting.
And I always worry about consistency. Does the first paper receive the same level of attention as the last paper when I am several hours into grading?
Experimenting With AI as a Tool
This semester, I decided to experiment.
When people hear “grading with AI,” the immediate reaction is often negative. In many cases, that reaction is justified. There are practices that should not be happening.
For example, faculty should not upload a student paper and a rubric into an AI tool and ask it to grade the assignment. If that is the model, we do not need instructors.
Instead, I am using AI in a different way.
I am using it as a transcriptionist.
Here is my process:
I open the student’s assignment in one browser window.
In a second window, I open ChatGPT and activate the transcription feature.
As I review the paper, I speak my feedback aloud.
I respond to the student conversationally:
“You described X accurately. That demonstrates a strong understanding of the theory.”
“You did not fully develop this concept. Let me explain where the gap is.”
I explain strengths. I explain weaknesses. I offer examples. I connect ideas across the paper. I talk through patterns I notice from the beginning to the end.
When I finish, I end the transcription and ask ChatGPT to organize my spoken feedback into a clear, professional written response for the student.
The result has typically been 500 to 700 words of structured, comprehensive feedback.
What Has Changed
Three things have improved immediately.
First, efficiency. I type slowly. Speaking is significantly faster for me. I no longer spend excessive time revising sentences for clarity and grammar. The tool handles the mechanical refinement.
Second, cognitive flow. When speaking, I think more holistically. I naturally connect themes across sections of the paper. If I notice repetition of an error in two places, I can ask the system to consolidate those ideas into a unified comment.
Third, consistency. Because the process is less physically and mentally draining, I feel more confident that the tenth paper receives the same level of attention as the first.
Student Response So Far
I teach in a graduate program, so expectations are high. After four weeks of using this method, I asked students directly whether the feedback felt useful, relevant, and thorough.
The responses have been positive.
Students report that the feedback is detailed and actionable. They see specific reinforcement of what they did well and clear explanations of what needs improvement.
From their perspective, nothing has been automated. They are still receiving my analysis, my interpretation, and my academic judgment. The difference is simply the medium through which that feedback is produced.
Boundaries Matter
This approach does not remove faculty from the grading process. It amplifies faculty voice.
The instructor still reads every paper.
The instructor still makes every evaluative judgment.
The instructor still determines the score.
The AI does not decide. It organizes.
That distinction is critical.
In my view, the ethical line is crossed when AI replaces human judgment. It is not crossed when AI enhances human efficiency.
Where This Goes Next
I plan to continue the experiment for the remainder of the semester.
There are additional questions worth exploring:
• How does this method scale across different assignment types?
• Does it work as effectively for shorter assignments?
• Can it be integrated into feedback cycles with revisions?
I remain cautious. I remain reflective. But I also remain open.
If we expect faculty to provide meaningful, individualized feedback in online environments, we must acknowledge the workload involved. Ignoring the strain does not solve it.
Used thoughtfully, AI can help us protect what matters most: instructional quality.
The goal is not to grade faster.
The goal is to teach better.