AI-Generated Slides: Are They Good? Can Students Tell?

Synthesis: This study evaluated five generative AI tools for creating instructional slides from instructor-authored course notes: NotebookLM, Claude, M365 Copilot, Cursor, and Claude Code. Educators assessed slides for accuracy, completeness, and pedagogical soundness.

• Coding assistants (Cursor, Claude Code) produced the best slides — most accurate, complete, and pedagogically sound.
• Students rated AI-generated slides as similar in quality to instructor-created slides.
• Students could not reliably identify which slides were AI-generated.
• A negative correlation emerged: high quality ratings were associated with lower "AI-generated" guesses, suggesting students associate poor quality with AI origin even when AI slides are good.

What this means for practice

• Instructors. Draft slides with a coding assistant rather than a dedicated education tool or a general-purpose LLM: Cursor and Claude Code produced the most accurate, complete, and pedagogically sound segments from the instructor's existing e-book materials, suggesting that the Scaffolding and structured output formats of coding tools carry over to instructional design.
• Instructors. Budget iteration time for visual consistency: matching AI-generated decks to the instructor's own style required repeated prompting, and styling remained inconsistent even after the prompts were refined.
• Faculty developers. Coach colleagues to demonstrate quality alongside disclosure — students associated low quality with AI origin (higher-rated segments drew fewer "AI-generated" guesses), so transparency about AI use can backfire unless the material is visibly good.
• Instructors. Ask students to judge the slides specifically; in this study the peer-instruction ratings could also reflect live coding and verbal explanations delivered in the same segment.
• Researchers. Run a controlled comparison of AI- and instructor-created slides with an actual learning measure, since this case study deliberately did not measure effects on learning.

Limitations

• Case study rather than research design: the authors state their goal was to report holistic initial experiences, not to conduct a research study.
• One course, one instructor, one term: a Web Software Development instance in fall 2025 with 88 students enrolled and 51 attending the first lecture, across five 2-hour lectures that mixed 4 Cursor-generated and 6 Claude Code-generated slide segments with instructor-made slides.
• Ratings were voluntary and uneven: participation did not affect course credit, individual students contributed different numbers of ratings, and the segment ratings may capture the live coding and verbal explanations around the slides.
• No learning-effect or transfer evidence: the design cannot show whether AI-created slides change learning or whether the pattern holds in another instructor's course.

Citation

Leinonen, J., Zhang, L., & Hellas, A. (2026). AI-generated slides: Are they good? Can students tell? Proceedings of the Western Canada Conference on Computing Education (WCCCE 2026).