Evidence-Based Research on Rori
At Rori, we are committed to evidence-based learning. Our goal is not just to build engaging technology, but to demonstrably improve math learning outcomes through rigorous research and continuous iteration.
Proven Impact in Classrooms
Early-stage research from the University of Oxford and J-PAL found that students who used Rori regularly achieved “markedly higher scores” in maths.
The study, conducted over 8 months, included 1,000 students in Grades 3–9 across 11 schools in Ghana.
Students who received two 30-minute sessions with Rori per week, in addition to their normal math lessons, showed:
• Significantly higher math performance
• An effect size of 0.36 suggests that using Rori may be comparable to roughly an additional year of learning gains.
• Strong engagement across grade levels
“While the results should be interpreted judiciously, as they only report on year one of the intervention, and future research is necessary to better understand which conditions are necessary for successful implementation, they do suggest that chat-based tutoring solutions leveraging artificial intelligence could offer a cost-effective approach to enhancing learning outcomes for millions of students globally.”
What Researchers Said
Beyond Test Scores: How Rori Contributes to Student Learning And Educational Research
In addition to large-scale impact evaluations, Rori is supported by a growing body of research exploring how AI tutoring improves learning quality, safety, and personalization.
-
Levonian et al. (2023) explore how retrieval-augmented generation (RAG) improves the quality and factual accuracy of math answers by grounding AI responses in trusted curriculum content.
-
Levonian et al. (2025) provide guidance for building safe, structured, and factually grounded generative tutoring systems, with practical recommendations for moderation and responsible AI use.
-
Henkel et al. (2025) introduce the AMMORE dataset (53,000 student answers) and demonstrate how chain-of-thought prompting improves the accuracy of automated grading for open-response math questions.
-
Vanacore et al. (2025) analyze millions of learner interactions to understand when students shift to easier content and how motivation and frustration shape learning decisions.
-
Moreau-Pernet et al. (2025) apply machine learning to predict when students are working on content that is too easy or too difficult, enabling proactive personalisation.
-
Zhang et al. (2025) find that slightly higher mastery thresholds (0.98 vs 0.95) lead to better long-term learning outcomes and smoother progression across lessons.
-
Cutler, Levonian & Christie (under review) show how improved intent detection in chat-based tutoring systems reduces user frustration and enables more flexible, student-centred learning pathways.
What This Means for Learners
Overall, the research indicates that Rori:
may improve measurable math performance
adapts to learning difficulty in real time
encourages persistence and resilience
provides safe, structured AI interactions
may support long-term skill growth
works effectively at scale but with low cost
delivers lessons designed by experts
gives back to the research community
Keeps improving thanks to new research