GenAI tools, such as ChatGPT, DeepSeek, and Perplexity, are transforming education. As a teacher myself, I have used them for material design, assessments, and lesson planning. They save me time, allowing me to focus on grading and refining my teaching. Without a doubt, they’re a game changer for teaching. But are they just as transformative for learning?
Many studies highlight benefits like personalized learning and immediate feedback (Joshi et al., 2021), which intuitively seem to improve students’ learning. However, do students really use GenAI for these purposes? Many teachers worry their students are overrelying on GenAI tools, which is actually making them lazy and dependent on those tools (Huang et al., 2024), leading them to avoid the hard tasks of thinking and memorizing, weakening their cognitive skills.
Faced with these opposing views, I set out to explore whether GenAI tools actually improve learning outcomes, as measured by students’ grades. This was challenging, as most research focuses on perceptions rather than controlled studies, comparing GenAI use tools with actual students’ outcomes.
As it turns out, the limited research available suggests that GenAI does improve learning outcomes; however, its impact is smaller than expected. Studies suggest a significant but small effect (Zhu et al, 2025), with variations depending on the educational level. Some research shows students at primary level benefit the most (Zhu et al., 2025), while others argue university students most benefit possibly due to the intellectual demand of higher education (Wang & Zhao, 2024). Despite these differences,both studies agree that GenAI has a positive effect on students’ learning outcomes, regardless of the tools used.
This struck me as odd, as it contradicted my own experience, where AI seemed to hinder learning. It made me wonder - what conditions are necessary to make GenAI improve students’ outcomes? Liam et al.’s (2023) study (an amazing paper, BTW) provided some interesting insights to respond to this question. Their research found that cognitive engagement and self-efficacy serve as mediators between students, GenAI, and learning outcomes.
It all made sense! The more cognitively engaged students were with GenAI, the more they would learn. This means, the more students use AI to think more actively by solving problems and interacting with information, the more they will learn, which, as a result, will improve their academic outcomes. This goes hand-in-hand with students’ sense of self-efficacy, their belief in their own learning abilities. When students use AI as a means to solve a problem, their self-efficacy increases, which, again, results in higher learning outcomes. Interestingly, cognitive engagement had an even stronger effect than self-efficacy alone. These insights are phenomenal, as they allow us to create actionable steps to successfully integrate AI in the classroom.
Given its potential benefits on students’ outcomes, we can’t ignore GenAI. Students are using it and banning it could trigger the streisand effect, which suggest that prohibiting something could increase its appeal (Jansen & Martin, 2015). Instead, teachers’ focus should shift away from punishment to accountability (Lim et al., 2023), allowing students to acknowledge AI without the fear of penalties.
This shift in mindset calls for an explicit AI-use policy in the classroom, outlining clear steps, prompts, examples of correct use and accountability measures. This policy should also be modelled in the classroom by the teacher showing they use it for their own learning, and even in materials or assessments.
Additionally, we must rethink classroom activities. To promote cognitive engagement, the activities should require problems-solving and creative solutions, encouraging students to use AI interactively rather than passively. It is this interaction which will lead to improved learning outcomes while building confidence in their abilities.
Initially, I was skeptical about AI’s role in learning, afraid it would hinder students’ outcomes. However, I now see its potential and embrace the challenge to help my students to use it wisely.
Huang, Z., Peng, N., Han, Y., Huang, B., & Qi, J. (2024). The role analysis of GenAI for college students—Evidence from China. In Li, E.Y. et al. (Eds.) Proceedings of The International Conference on Electronic Business, Volume 23 (pp. 469-480). ICEB’24, Zhuhai, China, October 24-28, 2024.
Joshi, S., Rambola, R. K., & Churi, P. (2021). Evaluating artificial intelligence in education for next generation. In Journal of Physics: Conference Series (Vol. 1714, No. 1, p. 012039). IOP Publishing.
Zhu, Y., Liu, Q., & Zhao, L. (2025). Exploring the impact of generative artificial intelligence on students’ learning outcomes: a meta-analysis. Education and Information Technologies, 1-29.
Wang, L., & Zhao, M. (2024). Can artificial intelligence technology promote the improvement of student learning outcomes? Meta-analysis based on 50 experimental and quasi-experimental studies. Proceedings of the 3rd International Conference on Educational Innovation and Multimedia Technology (EIMT 2024), Wuhan, China. https://doi.org/10.4108/eai.29-3-2024.2347685
Liang, J., Wang, L., Luo, J., Yan, Y., & Fan, C. (2023). The relationship between student interaction with generative artificial intelligence and learning achievement: Serial mediating roles of self-efficacy and cognitive engagement. Frontiers in Psychology, 14. https://doi.org/10.3389/fpsyg.2023.1285392
Lim, W. M., Gunasekara, A., Pallant, J. L., Pallant, J. I., & Pechenkina, E. (2023). Generative AI and the future of education: Ragnarök or reformation? A paradoxical perspective from management educators. The international journal of management education, 21(2), 100790.
Jansen, S. C., & Martin, B. (2015). The Streisand effect and censorship backfire. International Journal of Communication, 9, 16.