AI Tutors in Schools

As of now

A World Bank pilot in Edo State, Nigeria (June-July 2024), found about 0.3 standard deviations of learning gain in six weeks from an after-school programme with a generative AI tutor and teacher guidance; the authors equate this with nearly two years of typical learning and call the result preliminary. A Harvard randomised crossover trial (194 students, published June 2025) found a GPT-4 tutor gave more than twice the median gain of an in-class active-learning lesson, in 49 instead of 60 minutes. A PNAS field experiment with nearly 1,000 Turkish high-school pupils (2025) found a plain chatbot raised practice scores but lowered later exam grades by 17%, while a version with teaching safeguards largely avoided the drop. In an exploratory trial by Google's LearnLM team and Eedi (165 pupils in five UK schools, December 2025), pupils tutored by the AI under human supervision solved new problems 66.2% of the time against 60.7% with human tutors alone; tutors approved 76.4% of the AI's drafts with at most minimal edits. A larger 1,525-pupil UK trial of a tutor reviewed by human tutors (10 schools, 12 weeks, outcomes due 2026) had not reported at the time of writing.