AI tutors and learning outcomes
Six weeks of teacher-guided AI tutoring improved English results in Nigeria
In short: Senior secondary students in Benin City, Nigeria, who attended twelve after-school English sessions using Microsoft Copilot, with teachers guiding and suggesting prompts, scored about 0.3 standard deviations higher on a final assessment than students who did not. They also did better on the regular end-of-term English exam, which covered more than the programme did.
| Paper | From Chalkboards to Chatbots: Evaluating the Impact of Generative AI on Learning Outcomes in Nigeria |
| Authors | Martin De Simone, Federico Tiberti, Maria Barron Rodriguez, Federico Manolio, Wuraola Mosuro, Eliot Jolomi Dikoru |
| Published | 20 May 2025 · World Bank Policy Research Working Paper 11125 |
| Study type | Randomized controlled trial · Working paper |
| What we read | Abstract and parts of the full text read. We read the abstract, the working paper's introduction and results sections, and the World Bank's summary of the programme. |
| Summarized |
What they did
Students who signed up were randomly selected to take part or not. Over six weeks in June–July 2024, participants attended twelve 90-minute after-school sessions in school computer labs, usually in pairs. Teachers introduced a topic, students worked with Copilot (GPT-4 at the time) using suggested prompts designed to make it act as a tutor rather than an answer machine, and sessions ended with reflection.
Who took part
First-year senior secondary students (around age 15) in nine public schools in Benin City, Edo State. 657 were assigned to the programme and 671 to control; 759 sat the final assessment.
What they found
- Final assessment: +0.31 SD overall (English, digital skills and AI knowledge combined).
- English skills, the main outcome: about +0.24 SD (the abstract reports 0.23).
- Regular third-term English exam, covering the whole year: +0.21 SD.
- Benefits appeared across ability levels and were largest for girls and for students with higher prior results.
Effect size: +0.31 SD overall; about +0.24 SD on English; +0.21 SD on the term exam (intention-to-treat).
Limitations
- The control group had no alternative programme, so extra structured study time with teachers is part of the effect.
- Only about 57% of randomized students sat the final assessment.
- Six weeks, with early sessions partly spent setting up accounts and learning to use computers.
- Power and internet cuts disrupted sessions.
- A working paper at the time of summary.
What this means for learners and parents
- AI tutoring worked here as part of a structured programme with teachers and well-designed prompts, not as a standalone app.
- Prompts that make the AI act as a tutor matter; the same tool can also just produce answers.
- Results from low-resource settings with no alternative tutoring may be larger than in homes that already have plenty of support.
Put it into practice (free tools)
Citation
Martin De Simone, Federico Tiberti, Maria Barron Rodriguez, Federico Manolio, Wuraola Mosuro, Eliot Jolomi Dikoru (2025). From Chalkboards to Chatbots: Evaluating the Impact of Generative AI on Learning Outcomes in Nigeria. World Bank Policy Research Working Paper 11125. https://doi.org/10.1596/1813-9450-11125
Related summaries
- Unrestricted ChatGPT help raised practice scores but lowered exam scores · Hamsa Bastani et al., 2025 · Preregistered randomized controlled trial
- A carefully designed AI tutor beat an active-learning physics class on a short-term test · Greg Kestin et al., 2025 · Randomized controlled trial with a crossover design
- Supervised AI tutoring matched human tutors in UK maths classrooms · LearnLM Team (Google) et al., 2025 · Exploratory randomized controlled trial
- An AI coach for human tutors raised topic mastery, most for weaker tutors · Rose Wang et al., 2024 · Preregistered randomized controlled trial