Can Students Tell AI-Generated Lecture Slides from Human-Made Ones? Their Guessing Heuristic Is Wrong

Forum topic · 小凯 · 2026-05-18

Summary

A study by Leinonen, Zhang, and Hellas had five AI tools (NotebookLM, Claude, M365 Copilot, Cursor, and Claude Code) generate lecture slides from instructor-written notes. Education experts evaluated the output quality, and the slides were then shown in real classrooms where students rated them and tried to identify which ones were AI-generated. Key findings: coding assistants (Cursor, Claude Code) produced more accurate and complete slides than general-purpose chatbots, likely because slide generation is a structured-document task suited to their strengths. In the classroom, students rated AI-generated slides no differently from instructor-made ones and could barely distinguish them, performing only slightly better than chance. Most strikingly, students used a quality heuristic: higher-rated slides were assumed to be human-made, lower-rated ones attributed to AI—a strategy that fails once AI output reaches human-level quality. Open questions include the subject matter used, the quality baseline of instructor slides, and the weighting of evaluation criteria.

Every university instructor faces the same reality: there is never enough time to prepare course slides. Leinonen, Zhang, and Hellas did something about it — they had five different AI tools (NotebookLM, Claude, M365 Copilot, Cursor, Claude Code) generate lecture slides from instructor-written course notes, had education experts assess the quality, and then put the slides into real classrooms where students rated them and tried to guess which ones were AI-made.

Differences Between AI Tools

General-purpose LLMs (Claude, Copilot) produced slides that were less accurate and complete than the two coding assistants (Cursor, Claude Code). The reason: coding assistants are better at handling structured output. Generating slides is essentially a structured document generation task, and coding assistants work daily with tightly formatted code blocks. It is a subtle observation — AI coding ability transferred to slide creation, while general chat models underperformed.

The Key Classroom Experiment

Students rated AI-generated slides no differently from slides made by instructors themselves. More critically: students could not reliably identify which slides were AI-generated. Accuracy was only slightly better than chance.

The Most Interesting Finding: A Negative Correlation

The higher a student rated a slide, the more likely they were to believe it was human-made. The lower the rating, the more likely they said "this must be AI-written." Students were not identifying "AI style" cues — they were inferring origin from quality: high quality → human-made; poor quality → AI-made. This heuristic completely breaks down once AI-generated slides reach human-level quality.

What Remains Unclear

Which course and subject the slides came from — AI slide-generation quality may vary considerably across disciplines.
The quality baseline of the "instructor-made" slides — ordinary PowerPoint-template slides, or carefully designed teaching materials?
How the evaluation criteria (accuracy, completeness, pedagogical soundness) were weighted.

---

References

1. Leinonen, J., Zhang, L., & Hellas, A. (2026). *AI-Generated Slides: Are They Good? Can Students Tell?* arXiv:2605.13532 [cs.AI]. 2. Mollick, E., & Mollick, L. (2023). *Using AI to Implement Effective Teaching Strategies in Classrooms*. Wharton School. 3. Hellas, A., et al. (2018). *Predicting Academic Performance: A Systematic Literature Review*. ITiCSE.