Google DeepMind: Gemini AI Tutor Gave Sierra Leone Students 1.5 Years of Progress in 8 Weeks
Google DeepMind has published the results of a large-scale experiment in Sierra Leone: 1,763 students studied math with Gemini for eight weeks and gained 1.5–2.5 school years of progress. In 91% of conversations, students used the AI to understand the material rather than to get ready-made answers. Gemini asked clarifying questions in 76% of its responses and gave out solutions in only 2% of cases. 69% of students met the usage targets, compared with a typical 5% for educational technology.
AI-processed from DeepMind Blog; edited by Hamidun News
Google DeepMind, together with Fab AI and Sierra Leone's Ministry of Education, has completed a large-scale randomized controlled trial (RCT): 1,763 secondary school students studied math for eight weeks using Guided Learning in Gemini. The trial is part of DeepMind's global program to study the real impact of AI on learning.
How the experiment was set up
The study covered 12 schools in the Port Loko district. The involvement of the government ministry of education provided access to real classrooms and official policy-level support. The study was pre-registered, which makes it one of the most rigorous in EdTech.
Students did not simply ask Gemini for answers: the Guided Learning feature is specifically tuned to nudge them toward thinking for themselves. The approach builds on years of research from the LearnLM program.
An analysis of 113,000+ conversations showed:
- In 91.4% of conversations, students used Gemini to understand concepts rather than to get ready-made answers
- In 76% of its messages, Gemini responded with clarifying questions, a "Socratic" approach
- The AI gave out ready-made solutions in only 2% of cases
- The key design principle: the cognitive load stays with the student and is not taken over by the AI
The results in numbers
Students who used Guided Learning showed a gain of +0.258 standard deviations in math relative to the control group. In practical terms, that is 1.2–1.7 school years of progress in eight weeks.
In classes where teachers used Gemini in roughly half of their lessons (the target was 12 hours over the period), the result was higher still: 1.8–2.5 school years.
Meanwhile, 69% of students met or exceeded the usage targets. That contrasts sharply with the "five percent problem": the share of users who typically stick with edtech tools voluntarily. In this study, the barrier was cleared by more than a factor of ten.
Over time, students' conversations with Gemini became more focused on learning: they looked for hints less often and asked conceptual questions more often.
"We need to study the results of our innovations rigorously.
I'm glad we now have solid evidence that carefully designed AI can improve learning outcomes in support of our teachers," said Conrad Sackey, Sierra Leone's Minister of Basic and Senior Secondary Education.
The role of teachers
The experiment was built as a partnership, not a replacement. Teachers designed lessons, set goals and led class discussions. Gemini acted as a personal tutor for each pair of students while the teacher moved around the room and supported those who were stuck.
In focus groups, teachers noted that working with the tool also affected their own preparation: using Gemini to develop lessons, they found new ways to explain long-familiar topics, such as fractions. Many described a change in their role, from "lecturers" to "facilitators."
DeepMind and Fab AI are releasing a methodological guide with protocols for other schools.
What this means
This is one of the first large-scale randomized studies of generative AI in education in the Global South. It suggests that a well-designed AI tutor does not replace teachers and does not open a loophole for cheating. It amplifies what teachers already do and helps students think for themselves.
For the EdTech industry, it is an argument for "Socratic" design: ask questions instead of handing out answers.
- Nov 21, 2026A swimming team captain handed ChatGPT three scheduled tasks and cut her routine work by a multiple
- Nov 21, 2026MIRA Neural Network Renders Rocket League at 20 FPS, but Ping Hits 460 ms
- Nov 21, 2026Google built a closed copy of the internet with fake banks to train Chrome's AI agents
- Nov 21, 2026MLPerf Training Adds Nvidia Rubin and TPU Ironwood in First Appearance: 153 Systems
Key Questions Answered
How the experiment was set up?
The study covered 12 schools in the Port Loko district. The involvement of the government ministry of education provided access to real classrooms and official policy-level support. The study was pre-registered, which makes it one of the most rigorous in EdTech. Students did not simply ask Gemini for answers: the Guided Learning feature is specifically tuned to nudge them toward thinking for themselves. The approach builds on years of research from the LearnLM program.
What are the key details on The results in numbers?
Students who used Guided Learning showed a gain of +0.258 standard deviations in math relative to the control group. In practical terms, that is 1.2–1.7 school years of progress in eight weeks. In classes where teachers used Gemini in roughly half of their lessons (the target was 12 hours over the period), the result was higher still: 1.8–2.5 school years. Meanwhile, 69% of students met or exceeded the usage targets.
What are the key details on The role of teachers?
The experiment was built as a partnership, not a replacement. Teachers designed lessons, set goals and led class discussions. Gemini acted as a personal tutor for each pair of students while the teacher moved around the room and supported those who were stuck. In focus groups, teachers noted that working with the tool also affected their own preparation: using Gemini to develop lessons, they found new ways to explain long-familiar topics, such as fractions.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.