The results of this pre-registered study suggest that AI can be a powerful pedagogical partner—not replacing teachers, but extending their reach. This study is part of ours constant effort build a global evidence base on the impact of artificial intelligence on teaching and learning.
Beyond the Answer Engine: Protecting Critical Thinking
A common concern is that generative AI could become a shortcut for students, potentially bypassing the demanding but necessary cognitive effort required for deeper learning. Guided learning aims to solve this problem: it is the result of many years of research and work in our company Find out LM’s efforts be pedagogically grounded and specifically tailored to prioritize building understanding before providing direct answers.
Data from Sierra Leone suggests this approach is working. An analysis of over 113,000 interactions exchanged during our trial period found that in 91.4% of conversations, students used the tool to build conceptual understanding rather than simply looking for solutions. Gemini responded with follow-up questions in 76% of its communications, but provided direct solutions in just 2% of cases. This “Socratic” interaction ensures that the student is left with the cognitive bulky lifting.
Teacher-led intervention
The success of this study was based on a partnership between AI and teachers that kept teachers at the center of the experience. Teachers designed lessons, set goals, and led classroom discussions that fueled learning.
In focus groups, teachers said Gemini also supported their own professional development. Using the lesson preparation tool, they discovered novel ways to explain familiar topics such as fractions. Many described a shift from “lecturers” to “facilitators” who moved around the classroom to support pairs of students in their own learning journeys.
To facilitate others implement similar programs, we are sharing the file teacher training guide with materials created in collaboration with Fab AI, including the specific protocols used in this study.
Impact measurement
The quantitative results were significant. Students using guided learning saw a +0.258 raise in standard deviations in their math scores compared to the control group. In practice, this equates to approximately 1.2 to 1.7 years of typical learning progress achieved during the eight-week trial period.
Students in classrooms where their teachers incorporated Gemini into about half of their lessons to meet the 12-hour target during the trial period saw even greater progress—progress of approximately 1.8 to 2.5 years. Engagement was also exceptionally high: 69% of students met or exceeded usage goals, well above the five percent typical for voluntary learning technologies (known as “The five percent problemThis means that students were not only engaged, but also more willing to come to classes.
Beyond the numbers, we also saw a profound change in behavior. Students reported enjoying mathematics more and actively engaging in learning beyond regular instruction. Most importantly, over time, their conversations and questions became more learning-oriented and moved towards developing skills rather than looking for immediate solutions. In particular, the number of skill-building questions increased to 90% in the last week – from 68% in the first week – while the number of solution-finding questions dropped from 25% to 10%, proving that students not only wanted the answer, but also understood how they got there.
The path forward
While these results are promising, they also highlight the challenge of the “achievement gap.” Although most students benefited, the greatest gains were among those who entered the study with better math skills. This highlights an important need: to offer the tools that provide the greatest benefit to the students who need them most.
Looking ahead, we plan to expand this research to other countries and delve deeper into areas such as metacognition and relational intelligence to capture a more holistic view that explores the nuanced complexities of learning. By combining the relational foundations of the teacher-led student classroom with the personalized, empowering capabilities of artificial intelligence, we can help ensure that technology is a bridge to meaningful learning opportunities for all.
1 We also received support from Google.org and the Gates Foundation to conduct the study. EducAid, laterite AND Oxford MeasurEd he also cooperated with us.