Introduction
Generative AI in education is touted as a game-changer. From personalized tutoring to automated assessments, the promise of AI transforming learning outcomes seems almost limitless.
Yet, as TopSchool.ai has discovered through practical implementation and examining emerging research, the reality is far more nuanced. While AI does indeed hold transformative potential, its impact varies significantly across disciplines and educational contexts.
In some cases, uncontrolled and aimless use of generative AI can harm learning by leading students to shortcut directly to answers, bypassing critical thinking and problem-solving processes. This is particularly problematic in subjects that require step-by-step reasoning, such as math and coding.
The role of AI in education must be reimagined, not as a stand-alone solution but as an augmentation to the teacher’s role. At TopSchool.ai, we have come to understand that the real power of AI in education is unlocked when it is thoughtfully designed, with human educators guiding the process, leveraging classic UI/UX principles, and configuring the underlying AI technology to align with educational objectives.
The Promise of Generative AI in Education
While generative AI tools like ChatGPT and Claude are often promoted as revolutionary educational aids, their general-purpose nature poses inherent risks when used by young learners.
These systems are not specifically designed with pedagogical frameworks in mind, which means they can inadvertently encourage surface-level engagement with content rather than deeper learning. Studies confirm that generative AI can have positive effects when used under controlled conditions and with teacher guidance.
A meta-analysis of AI in education found a significant positive effect size (Hedges’ g = 0.86), particularly when AI tools like chatbots were structured to support learning, yielding an impressive effect size of 1.02 for improving student outcomes (Zhang et al., 2025).
To put this in real terms, this effect size translates to approximately a full grade level improvement, suggesting that when used correctly, AI can help students perform markedly better in assessments.
However, the effectiveness of generative AI is not universal. The positive outcomes reported tend to be most prominent when AI is used as an interactive learning companion rather than as a tool for simply completing tasks.
For example, a study on an AI-powered math tutor in Ghana (Henkel et al., 2024) showed that students who engaged thoughtfully with the tutor saw learning gains equivalent to an additional year of schooling (effect size = 0.36).
However, indiscriminate use, particularly in more conceptual disciplines, can reduce learning efficacy as students rely on AI to bypass deeper cognitive processing.
The Pitfalls of Unstructured AI Use
In educational practice, the very act of learning involves moving towards answers through the synthesis and analysis of concepts. AI systems designed for young learners must slow the process down to encourage deeper engagement.
Studies involving ChatGPT in coding classes revealed that while some students benefited from using AI to explore explanations, others who relied on it to complete exercises without effort saw diminished outcomes (Lehmann et al., 2024).
This reliance on AI for quick answers can lead to a false sense of mastery, where students believe they have understood concepts simply because they obtained the correct result. Similarly, a high school experiment using GPT-4-based math tutors demonstrated that students using a basic GPT interface performed worse in subsequent exams compared to those without AI assistance (Bastani et al., 2024).
This negative impact was mitigated when the AI was designed to guide the problem-solving process rather than provide direct answers, emphasizing the importance of thoughtful AI configuration.
The Evolving Role of Teachers
The role of the human educator is not only preserved but becomes even more critical in an AI-enhanced educational environment. At TopSchool.ai, we recognize that educators must be equipped with the skills to effectively integrate AI into their teaching practices.
This means professional development support must evolve alongside technological advancements. Teachers are not just facilitators but architects of learning experiences, leveraging AI to enhance student engagement and understanding.
As AI becomes more integrated, their role will shift, requiring new skills and pedagogical approaches.
Conclusion
Generative AI can indeed revolutionize education, but only if integrated responsibly. The key lies in combining thoughtful AI design with active teacher involvement and robust professional development.
At TopSchool.ai, we envision a future where AI and human educators work in tandem, combining the insights and data-driven capabilities of technology with the empathy, experience, and judgment of teachers.
This model of ‘co-intelligence’ , where AI amplifies proven teaching practices while allowing room for innovation, represents the most promising path forward. This collaborative approach will ensure that we not only preserve the human element in education but enhance it, preparing students to thrive in a world shaped by both human and artificial intelligence.
To learn more about the findings of these meta analyses, find the original papers below.
Original Papers
Meta-Analysis of Artificial Intelligence in Education, Canadian Center of Science and Education, Higher Education Studies, March 2025 — https://www.researchgate.net/publication/390092849_Meta-Analysis_of_Artificial_Intelligence_in_Education
Effective and Scalable Math Support: Evidence on the Impact of an AI- Tutor on Math Achievement in Ghana, Owen Henkel, Hannah Horne-Robinson, Nessie Kozhakhmetova, Amanda Lee — https://arxiv.org/abs/2402.09809
AI Meets the Classroom: When Do Large Language Models Harm Learning?
Matthias Lehmann, Philipp B. Cornelius, Fabian J. Sting https://arxiv.org/abs/2409.09047
Generative AI Can Harm Learning, The Wharton School Research Paper, Jul 2024, Hamsa Bastani, Osbert Bastani, Alp Sungu, Haosen Ge, Özge Kabakcı, Rei Mariman — https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4895486