Assessment scenario: Students in CHEM 1210 and CHEM 1220 complete pre-lecture assignments by submitting questions about concepts they find confusing, generating hundreds of written responses before class. At this scale, manually reviewing and synthesizing enough responses to identify patterns in student thinking is impractical. Learn how generative AI can help organize this formative assessment evidence so instructors can identify areas of confusion and use class time to address concepts that most need clarification.
Instructor: Matthew W. Stoltzfus
Department: Chemistry and Biochemistry
Course: CHEM 1210 & CHEM 1220 General Chemistry I & II
Contact: stoltzfus.5@osu.edu
Course Details
Format: In-person
Typical enrollment: 300+ students
Instructional staffing: One primary instructor
Learning context: Undergraduate students majoring in the sciences who are enrolled in a large-lecture gateway chemistry course
Expected Learning Outcomes
- Reflect on one's own understanding of course material during pre-class preparation
- Identify specific concepts, questions, and areas of confusion before class
- Use lecture time for clarification, sense-making, peer discussion, and problem-solving rather than first exposure to content
- Engage more actively with difficult ideas that emerge during preparation
Opportunity
Use AI to make pre-class student thinking more visible
Before each class, I ask students to complete assignments that introduce the core concepts they will encounter in lecture. As part of that preparation, students respond to a “muddiest point” prompt in which they identify questions, confusing concepts, difficult problems, or connections they would like the upcoming class to address.
After students complete the assigned textbook reading and tutorial problems, they respond to the following prompt in a graded Canvas survey quiz:
“Based on your reading/watching the lecture videos and completing the pre-lecture tutorial problems, type out three questions that you would like the upcoming lecture to answer. If you do not have any questions, then summarize the key concepts from the pre-lecture assignment without referring to the text or to your notes.”
This prompt gives me a valuable window into students’ minds before class begins. The challenge is scale. In a large-enrollment chemistry course, a single assignment can generate 300 or more written responses before each lecture. In the past, I could briefly review some of those submissions, but I could not realistically read and synthesize all of them before class. As a result, I often relied on my assumptions about what students would find difficult rather than evidence of what they were actually struggling with.
Change
Use AI to identify patterns in student thinking
Generative AI created an opportunity to make better use of formative assessment evidence I was already collecting. By helping me identify patterns across hundreds of student responses, AI makes it possible to listen more carefully to the class as a whole and use that information to emphasize the concepts students need clarified during lecture. Rather than changing the purpose of the assignment, AI made an existing teaching practice more practical, responsive, and useful at the scale of a large course.
The general workflow is to download an Excel spreadsheet of the student responses from the Canvas quiz, remove identifying information, and paste the responses into a Google Gemini chat. I then ask Gemini to group similar questions, identify recurring areas of confusion, and summarize the concepts students most frequently want addressed. I review the summary, return to individual student responses when needed, and decide which ideas should receive additional emphasis during lecture.
AI is not evaluating individual students or determining what I should teach. Instead, it helps me organize a large amount of formative assessment evidence so that I can make those decisions with a clearer picture of what students are actually thinking. Even when my lecture materials are already prepared, this process allows me to make small but meaningful adjustments by clarifying a concept, adding an example, revisiting a connection to earlier material, or incorporating an actual student question into class discussion.
In this way, AI allows me to scale a teaching practice that would be much easier in a small class: listening carefully to what students do not understand before class begins and using that information to make the lecture more responsive to their needs.
Reflection
Learn from patterns in student thinking over time
This process reinforced how easy it is for the “curse of knowledge” to shape my expectations about what students should find difficult. Reviewing patterns in their responses gives me a more direct view of where they are actually struggling and helps me make more informed decisions about how to use class time.
Moving forward, I would like to track recurring misconceptions across the semester and use those patterns to inform review materials, peer discussion, and targeted in-class questions. AI also creates the possibility of building an archive of student responses that can be revisited not only throughout a single semester, but from year to year. This could help me identify recurring patterns in student thinking that I may not remember on my own and compare whether the same concepts continue to create difficulty for different groups over time. I am also interested in comparing the pre-class questions students ask with their performance on later formative assessments to better understand whether the issues identified before class are being resolved.
Conversation
Q&A: AI, Assessment, and the Future of General Chemistry
In this extended conversation with Steven Brown, I discuss what I have learned from using AI to better understand student thinking and explore broader questions about assessment, critical thinking, active learning, and the future of general chemistry.