World Bank GPT-4 English Tutoring RCT — Two Academic Years of Learning in Six Weeks, Edo State, Nigeria

A World Bank randomised controlled trial (Education Global Department, with Mastercard Foundation support) tested GPT-4-based English tutoring (via Microsoft Copilot) in nine public secondary schools in Benin City, Edo State, Nigeria, in June–July 2024. Eight hundred first-year senior secondary students attended twice-weekly after-school sessions combining teacher-framed topics with AI-assisted practice.

To test whether short-term, AI-assisted after-school tutoring can improve English-language proficiency among under-resourced public secondary school students, and to assess transfer effects to other subjects and cost-effectiveness relative to human tutoring.

Randomised assignment of students to treatment (AI-tutoring) and control groups; twice-weekly after-school sessions over six weeks in which teachers framed a topic before students practised English with Microsoft Copilot (GPT-4); teachers acted as facilitators ('orchestra conductors') rather than direct instructors during AI practice.

Treatment students significantly outperformed controls on English proficiency (the primary outcome) and on AI knowledge/digital skills. The six-week intervention produced English learning gains equivalent to two full academic years of conventional schooling. Gains transferred to end-of-year exam performance in subjects not covered by the programme. A cost-effectiveness analysis found the approach substantially cheaper per learning unit than comparable human-tutor interventions.

The World Bank published a seven-lesson brief recommending scale-up conditional on teacher capacity-building, data-privacy safeguards, and reliable electricity/device access.