How to help your students use AI without losing the learning

It’s hard to believe ChatGPT arrived on the scene only 3 short years ago, yet generative AI has already become firmly entrenched into everyday academic life. Recent surveys paint a consistent picture: In 2024, the Digital Education Council reported that 86% of bachelor’s, master’s, and doctoral students used AI tools for their coursework—most often to brainstorm ideas, explain concepts, summarize readings, or prepare for assessments (Digital Education Council, 2024). Similarly, a 2025 Higher Education Policy Institute survey of undergraduates found that 92% used AI on assignments, with 58% using it to clarify difficult concepts and 48% to summarize articles (Freeman, 2025).

Without question, students are using AI. But how is all of this affecting their learning? The answer may depend less on whether students use AI and more on how they use it.

The ICAP framework can be used to organize student engagement into four levels: Interactive, constructive, active, and passive. The higher levels (interactive and constructive) are associated with the strongest learning gains because they require students to generate ideas, make connections, and work with concepts. In contrast, passive engagement (such as simply listening to a lecture) leads to far weaker understanding (Chi & Wylie, 2014).

AI threatens learning when it displaces these higher-level processes. If students outsource the thinking work of an assignment (crafting explanations or forming arguments), they can slide from constructive or interactive engagement into more passive modes. Skimming or lightly editing an AI-generated paragraph may feel efficient, but it bypasses the cognitive effort that drives deep, transferable understanding (Bauer et al., 2025). Over time, this pattern can weaken the development of domain knowledge, making it harder for students to recognize when an AI-generated explanation is biased, superficial, or inaccurate.

A recent field experiment illustrates these risks. In a study of nearly 1,000 high school math students, Bastani and colleagues (2025) compared two GPT-4 tutors: A standard ChatGPT-style interface (“GPT Base”) and a version prompted to support, rather than replace, learning (“GPT Tutor”). Both improved problem-solving performance during use (by 48% and 127%, respectively), but once all AI was removed and students were asked to perform a subsequent problem-solving task, students who used GPT Base scored 17% lower than a comparison group who never had AI access at all. This suggests that overreliance on AI hindered retention. What is more interesting is that these negative effects were largely avoided with the GPT Tutor condition, suggesting that the problem is not AI use itself, but in whether it is used to support or replace learning.

The authors of a December 2024 study in the British Journal of Educational Technology concluded that ChatGPT use may risk promoting “metacognitive laziness” by shifting thinking and problem solving away from students. In a lab experiment with 117 Chinese students writing English essays, participants used either ChatGPT, a human writing coach, a checklist-based toolkit, or no support (Fan et al., 2025). The ChatGPT group did produce the highest-quality essays, but showed no gains in learning, motivation, or interest. Participants in the ChatGPT condition also frequently copy-pasted text, suggesting far more passive engagement with learning materials.

Evidence seems to indicate that when students use AI constructively, learning can improve. A meta-analysis of 51 experimental or quasi-experimental studies using ChatGPT in a learning intervention (Wang & Fan, 2025) found large gains in participant learning (g=0.867) and moderate gains in perceived learning (g=0.456) and higher-order thinking (g=0.457). These results align with a December 2024 review of 69 studies (Deng et al., 2024), which concluded that ChatGPT use can enhance academic performance, motivation, and deeper cognitive engagement.

Individual classroom studies echo this pattern. In one international relations course, students used ChatGPT during debate exercises to generate immediate counterarguments and receive instant feedback (de la Puente et al., 2024). Compared to students in a non-AI control section, those who worked with AI, demonstrated significant improvements in both argumentation and critical thinking. In such cases, students weren’t handing their work over to the AI. Instead, they were interacting with it, critiquing it, and using it to sharpen their own reasoning. AI acted less like a ghostwriter and more like a thinking partner.

If poor AI use leads to reduced learning, students can adopt behaviors that keep cognitive effort where it belongs. A helpful guiding question for students is: “Is AI doing the learning for me, or is it helping me learn?”

To keep AI on the learning-supportive side, encourage your students to:

1. Start with their own attempt

Draft a rough answer, outline, or explanation before consulting AI. This ensures students have something to compare, critique, and revise.

2. Use AI to challenge, not replace, their thinking

Students can ask AI questions such as:

• What counterarguments am I missing?
• Point out weaknesses in my explanation.
• Give me alternative interpretations and help me weigh them.

3. Make their engagement visible

Ask students to include copies of their AI chats, summarize AI feedback in their own words, or explain what they agree or disagree with.

4. Be explicit about roles

Teach students that idea generation, sense-making, and judgment is their job, while AI’s role might be to offer examples, perspectives, and feedback that they can evaluate.

AI is already a central part of students’ academic lives. By anchoring AI use in the higher levels of the ICAP framework and keeping students cognitively engaged, we can help ensure that AI becomes a partner in learning, rather than a shortcut around it.

References

Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö., & Mariman, R. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. Proceedings of the National Academy of Sciences, 122(26), e2422633122. https://doi.org/10.1073/pnas.2422633122opens in new window

Bauer, E., Greiff, S., Graesser, A. C., Scheiter, K., & Sailer, M. (2025). Looking beyond the hype: Understanding the effects of AI on learning. Educational Psychology Review, 37(2), 45. https://doi.org/10.1007/s10648-025-10020-8opens in new window

Chi, M. T., & Wylie, R. (2014). The ICAP framework: Linking cognitive engagement to active learning outcomes. Educational psychologist, 49(4), 219-243. https://doi.org/10.1080/00461520.2014.965823opens in new window

de la Puente, M., Torres, J., Troncoso, A. L. B., Meza, Y. Y. H., & Carrascal, J. X. M. (2024). Investigating the use of ChatGPT as a tool for enhancing critical thinking and argumentation skills in international relations debates among undergraduate students. Smart Learning Environments, 11(1), 55. https://doi.org/10.1186/s40561-024-00347-0opens in new window

Deng, R., Jiang, M., Yu, X., Lu, Y., & Liu, S. (2024). Does ChatGPT enhance student learning? A systematic review and meta-analysis of experimental studies. Computers & Education, 227, 105224. https://doi.org/10.1016/j.compedu.2024.105224opens in new window

Digital Education Council. (2024, August 2). Digital Education Council Global AI Student Survey 2024. Digital Education Council Global AI Student Survey 2024opens in new window

Fan, Y., Tang, L., Le, H., Shen, K., Tan, S., Zhao, Y., Shen, Y., Li, X., & Gašević, D. (2025). Beware of metacognitive laziness: Effects of generative artificial intelligence on learning motivation, processes, and performance. British Journal of Educational Technology, 56(2), 489–530. https://doi.org/10.1111/bjet.13544opens in new window

Freeman, J. (2025). Student generative AI survey 2025. Higher Education Policy Institute. Retrieved from https://www.hepi.ac.uk/wp-content/uploads/2025/02/HEPI-Kortext-Student-Generative-AI-Survey-2025.pdfopens in new window

Wang, J., & Fan, W. (2025). The effect of ChatGPT on students’ learning performance, learning perception, and higher-order thinking: Insights from a meta-analysis. Humanities and Social Sciences Communications, 12, 621. https://doi.org/10.1057/s41599-025-04787-yopens in new window

About the author:

Joseph J. Slade, MS, is a PhD student at Oregon State University studying the intersection of AI, teaching, and learning. He serves as director of Project FAILSafe (Fostering AI Learning Safely), a collaboration between university departments of psychology and computer science that builds AI educational tools and tests their efficacy in laboratory and classroom settings. Joseph is also co-founder and president of AI Education Labs, Inc., an educational software company, as well as host of PsychSessions: AI, a podcast dedicated to the evolving relationship between education and AI.