evidoria

← Back to browse

Good practice

Gemini Guided Learning — A Pre-Registered RCT of AI Math Tutoring in Port Loko District, Sierra Leone

Sierra Leone · Port Loko · See the Sierra Leone profile

A pre-registered 8-week RCT of Google DeepMind's Gemini "Guided Learning" tutor across 1,763 students in 12 Port Loko schools found a +0.258 SD math gain (0.380 SD for high-usage students) — but higher-baseline students benefited most, an equity gap flagged by researchers.

1,763
Students randomised
48
Classrooms
12
Schools
+0.258 SD
Intent-to-treat effect size
0.380 SD
Treatment-on-treated effect (12h+ usage)
69%
Students meeting usage target
76%
Responses using scaffolding questions
2%
Responses giving direct answers
91.4%
Conversations pursuing conceptual understanding
113,000+
Logged interactions analysed

Details

Maturity
Pilot
Promoter
Google DeepMind
Period
October–December 2025
Keywords
generative AI, adaptive tutoring, mathematics, randomised controlled trial, junior secondary education

Context

Google DeepMind, with Fab AI, Sierra Leone's Ministry of Basic and Senior Secondary Education, Google.org and the Gates Foundation, ran a pre-registered RCT of the Gemini-based "Guided Learning" Socratic math tutor across 12 government-supported junior secondary schools in Port Loko District, Sierra Leone, from 6 October to 5 December 2025.

Results

1,763 Grade 7-8 students across 48 classrooms were randomised to Guided Learning or business-as-usual control; the intent-to-treat effect on externally validated math scores was +0.258 SD (0.380 SD for students meeting a 12-hour usage threshold); 69% of students met usage targets versus a ~5% typical baseline; however, higher-baseline students benefited more than lower-baseline students, widening the achievement gap.

Implementation

Implementation detail (cost, timeline, staffing, conditions for success) is not yet available for this practice.

Data sources

Where this practice's information was retrieved from, and when.

Similar practices you may find useful