What the Research Actually Shows — An Honest Assessment
Personalised learning powered by AI has been promised as the transformative future of education for approximately 15 years. These claims have attracted enormous investment in EdTech companies globally and generated widespread media coverage. They have also generated a body of peer-reviewed research that we can now evaluate honestly. The picture that emerges is both more positive than critics claim and more limited than advocates suggest.
What the Evidence Shows Actually Works
A 2024 meta-analysis of 47 randomised controlled trials examining AI-assisted learning interventions in K-12 and higher education settings found consistent, statistically significant improvements in knowledge retention and procedural skill development when AI was used for specific purposes: generating practice questions with immediate corrective feedback, providing adaptive difficulty (questions become harder when students succeed and easier when they fail), and delivering immediate, specific feedback on errors rather than delayed feedback at end-of-session.
The average effect size across all studies was approximately d = 0.42, which translates roughly to moving a student from the 50th percentile to the 66th percentile of academic performance relative to peers who did not use AI-assisted practice. This is a meaningful improvement — equivalent to approximately six additional months of learning over a two-year period — but not the transformational leap that marketing claims often suggest. The strongest effect sizes were found in mathematics and science procedural skills (d = 0.58 on average). The weakest effect sizes were found in essay writing and analytical argumentation (d = 0.22 on average) — domains where AI's ability to provide genuinely useful feedback is fundamentally limited.
One of the most persistent claims in EdTech is that AI can adapt to individual "learning styles" — visual, auditory, kinesthetic. This claim does not survive contact with the evidence. The learning styles hypothesis has been tested in hundreds of studies and consistently fails to show benefits when instruction is tailored to claimed learning style preferences. The categorisation itself is questionable — most people do not have a single stable learning style. AI systems that claim to adapt to learning styles are making a promise not supported by educational research. A student who is told by an AI that they are a "visual learner" and is subsequently given only visual content is being under-served, not personalised to.
The Genuine Limitations of Current AI in Education
Current AI systems have three fundamental limitations in educational contexts that enthusiastic proponents tend to downplay. First, AI cannot reliably evaluate conceptual understanding — only surface indicators of it. A student who has memorised the correct form of an answer without understanding the underlying concept will often produce responses that AI evaluates as demonstrating understanding. A skilled human teacher who knows the student, asks follow-up questions, and probes for genuine comprehension catches this quickly. Current AI does not.
Second, AI cannot provide the motivational, relational, and emotional support that is genuinely important to learning outcomes. Research consistently shows that the quality of the student-teacher relationship is one of the strongest predictors of academic achievement — more predictive than class size, homework quantity, or technology use. AI does not form genuine relationships. Students who need human connection for motivation cannot substitute AI for it.
Third, AI cannot teach values, judgement, and wisdom — the ultimate goals of education. Learning calculus is not the goal of mathematics education; developing mathematical thinking and logical reasoning is. AI can accelerate the acquisition of content knowledge, but it cannot develop the higher-order human capacities that content knowledge is supposed to build. Those capacities require teachers, communities, and genuine intellectual engagement with the world.
Experience evidence-based AI learning with Khypri AI — built on what the research shows works, honest about its limitations, and designed around your child's genuine educational development. Start free today.
The Right Way to Use AI in Your Child's Education
Given what the evidence shows about where AI-assisted learning works and where it does not, here is the practical framework for parents and students considering how to incorporate AI educational tools into their preparation. Use AI for practice and retrieval — the domain where the evidence of benefit is strongest. AI-generated questions, immediate feedback on answers, adaptive difficulty, and systematic tracking of which topics need more attention are the AI capabilities that translate most directly into improved examination performance.
Use AI to get unstuck — not to get answers. A student who has attempted a problem, failed to solve it, and then asks an AI to explain where they went wrong and why is engaging productively with the tool. A student who asks AI to solve their homework problems before attempting them themselves is undermining their own learning. The distinction is between AI as a patient explainer that responds to genuine confusion and AI as a shortcut that bypasses the cognitive effort that learning requires.
Do not use AI as a substitute for the teacher-student relationship. The evidence on the importance of that relationship to motivation, belonging, and long-term academic development is unambiguous. AI can supplement excellent teaching. It cannot replace it. A student who uses AI effectively alongside quality classroom teaching has a genuine advantage. A student who uses AI instead of engaging with their teachers and peers has made a poor trade.
Evaluate AI educational tools with the same rigour you would apply to any educational resource. Ask whether the tool's claims are supported by evidence, whether the content is curriculum-aligned and factually accurate, whether the safety features are appropriate for your child's age, and whether the data practices are compliant with DPDP Act 2023. The EdTech market is large, competitive, and poorly regulated. Not all tools that claim to be educational are genuinely educational, and not all tools that claim to be safe are genuinely safe.
Experience evidence-based AI learning with Khypri AI — built on what the research shows works, honest about its limitations, and designed around your child's genuine educational development. Start free today.