AI in Education: Does It Help Learning — or Hurt It?

It depends on how it is used. AI in education helps learning when it guides a student to think, and harms learning when it simply produces answers. The design of the tool, not the technology itself, decides the outcome.

This matters because the two outcomes look identical at first. In both cases the homework gets done, the grades often rise, and the child seems to be coping. The damage only shows up later, on a test the AI cannot sit.

So a useful way to judge AI in education is to ask one question: does the student still do the cognitive work? When the answer is yes, AI becomes a powerful tutor. When the answer is no, it becomes a shortcut that skips the part where learning actually happens.

The central risk of AI in education is simple: when students use AI to produce answers, their schoolwork usually improves while their learning declines. Several independent studies now document this gap between performance and understanding.

Performance up, learning flat

A peer-reviewed study in the British Journal of Educational Technology calls the effect metacognitive laziness. Students who wrote with ChatGPT improved their essay scores. However, their actual knowledge gain did not improve. Nor did their ability to transfer it to new problems (Fan et al., 2025).

The same pattern at scale

A 30-month study of more than 26,000 secondary students found the same split. After six months of AI use, homework scores rose by about 18%. Meanwhile, closed-book exam scores fell by roughly 20% (Strömberg, Lei & Wu, 2026).

The mechanism is simple

The cause is what researchers call homework outsourcing. When a child gets a finished answer with little effort, the AI does the thinking. Yet the thinking is exactly what builds learning. To understand the technology underneath, our guide to how AI actually learns is a useful companion.

The learning cost of AI in education grows over time and is easy to miss early on. On high-stakes exams, the penalty in the large secondary-school study reached 18–24% (Strömberg, Lei & Wu, 2026). It took nearly two years to fully appear.

It compounds quietly

Grades and homework look fine along the way, so families often notice nothing until a major test. As a result, short studies tend to underestimate the real cost. The gap widens the longer a student leans on the shortcut.

The crutch effect

A separate classroom experiment showed this directly. Students given a raw answer-giving chatbot did worse once it was taken away (Bastani et al., 2025). Early brain-imaging work points the same way, although that evidence is still preliminary (Kosmyna et al., 2025). In short, an answer-giving tool can leave a learner weaker than before, because it removed the struggle that builds skill.

The research is just as clear about the other side: AI in education helps learning when it is built on a few published teaching principles. Guide instead of answer. Make the learner do the thinking. Check understanding before moving on. Bring material back later so it sticks.

The positive evidence

A Harvard study tested an AI tutor designed around these principles, using expert-written scaffolds, step-by-step reasoning, and deliberately short prompts. As a result, students learned more than twice as much, in less time, than they did in an active-learning class (Kestin et al., 2025).

The lever is hints, not answers

A high-school maths experiment made the difference obvious. The version that handed over answers reduced learning. However, a second version — prompted to give hints rather than solutions — largely removed the harm (Bastani et al., 2025). In addition, an early UK classroom trial using a pedagogically-tuned model found students matched human-tutored peers and were more likely to solve novel problems afterwards — promising, though still early evidence (LearnLM Team, 2025). The lesson for AI in education is consistent: design decides everything.

Epivo is built to sit on the helpful side of this line: it leverages what AI does brilliantly — patience, availability, and personalised explanation — while refusing to do the one thing that breaks learning, which is handing over the answer.

Leveraging AI's real strengths

A good tutor is endlessly patient, available at any hour, and able to explain a concept ten different ways until it clicks. AI does all of this well, and Epivo uses it fully. That is the upside of AI in education, and there is no reason to give it up.

Designing out the pitfalls

The trap is letting the student skip the thinking. So Epivo guides instead of answering, asks the learner to explain their reasoning, and checks understanding before moving on. It keeps the cognitive work where it belongs — with the student — and brings ideas back later so they stick. For a fuller picture, see our overview for parents and how Epivo works for students.

In other words, Epivo is designed to capture the documented benefits of AI in education while avoiding the documented harms of an answer-giving tool.