RESEARCH EXPLAINED · 4 MIN READ

What does research actually say about AI and learning?

AI can make help more immediate. Whether that help improves learning depends on the way it is designed and used. Two published experiments show why that distinction matters.

Personalized tutoring can help.

Kestin and colleagues studied 194 students in a Harvard introductory physics course. Their randomized trial compared a purpose-built AI tutor with active classroom learning. Students learned more in less time with the tutor in that setting. The tutor followed specific teaching principles; the result does not apply automatically to every AI tool.

Source: Kestin et al. (2025), Scientific Reports · Read the study ↗

Faster answers do not guarantee stronger skills.

Bastani and colleagues ran a mathematics experiment with nearly 1,000 high-school students. AI access improved assisted practice performance, but students using an unrestricted version later performed worse when the AI was removed. A version designed around teacher-provided hints largely mitigated that negative effect.

Source: Bastani et al. (2025), PNAS · Read the study ↗

A practical way to use that evidence.

1. Try the question yourself before asking for help.
2. Read and check the explanation, including calculations and assumptions.
3. Close the answer and explain the reasoning in your own words.
4. Try a similar question without assistance.

These are practical suggestions drawn from the distinction between completing a task and learning a skill. They are not measured outcomes for GrindMate.