AI tutor that coaches students through problems

Eedi, a UK maths learning platform, and Google DeepMind ran an exploratory randomized controlled trial in the summer of 2025 across five UK secondary school classrooms, with 165 students and 17 expert tutors. Core tutoring replies were drafted by LearnLM, a Google DeepMind model built for pedagogy, and an expert human tutor reviewed and could edit every message before a student saw it.

Eedi reports the human AI team matched a human tutor alone on immediate mistake correction (93.0% versus 91.2%) and misconception resolution (95.4% versus 94.9%), and more than doubled the human tutor's effect on a knowledge transfer measure, boosting it by 10 percentage points against 4.5 percentage points for the human tutor alone, each compared with a static hint.

Tutors approved 82.3% of the AI's suggested messages with no or only minimal edits, and a full audit found zero harmful or risky messages and factual inaccuracies in 0.1% of messages.

Philip Oreopoulos and Nina Low ran a two year cluster randomized trial in 18 Tennessee middle schools in Hamilton County in the 2024/25 and 2025/26 school years, in which randomly assigned students used Khan Academy with its AI tutor Khanmigo, configured to coach rather than give answers, during existing daily remedial maths sessions. Assignment raised maths achievement by 1.3 national percentile ranks per term, similar to Khan Academy practice without AI.

Almost every student tried Khanmigo, but the median student messaged it on only a third of the days they practiced and in only 17 percent of the exercise sessions in which they made a mistake, and the messages students did send were mostly bare answers or clicks on suggested prompts. Chalkbeat reports that Khan Academy has since redesigned its interface to integrate Khanmigo better.

In June and July 2024 a World Bank team, working with the Edo State education authorities, ran a six week after school programme in Edo, Nigeria, in which 800 first year senior secondary students used Microsoft Copilot as a tutor, mainly to learn English, with teacher support: teachers introduced each session's topic, suggested prompts and mentored students as they worked with the tool.

In a randomized evaluation, participants outperformed their peers in English, AI knowledge and digital skills, and also did better in their end of year exams. The team reports learning gains of about 0.3 standard deviations, says the programme outperformed 80% of the interventions in a database of randomized evaluations in developing countries, and reports that the more sessions students attended, the greater their gains; girls, who started behind boys, seemed to gain even more.

Since spring 2023 Harvard's introductory computer science course CS50 has run the "CS50 Duck", an AI tutor built on the ChatGPT API that is deliberately less helpful than a general chatbot: it asks more questions than it answers and avoids giving outright solutions. The course combines prompting with its own code that tries to evaluate each reply before the student sees it and sometimes rejects or retries it, and added a "heart system" that limits questions per period after some students asked around 200 questions. David Malan calls it a net positive but acknowledges the Duck still sometimes returns code despite instructions not to.

Brockton Public Schools, a Massachusetts district of over 15,000 students, has used Carnegie Learning's MATHia, which Carnegie Learning describes as an AI powered math supplement that coaches students as they work, alongside its Middle School Math Solution and MATHstream since fall 2021.

Carnegie Learning reports that at Brockton's North Middle School, high MATHia usage helped drive a 400% increase in MCAS passing rates between 2021 and 2024, and that in 2023/24 students rated MATHia proficient met or exceeded expectations on the MCAS at a higher rate than the district and state averages.

District wide, the share of middle school students meeting or exceeding MCAS expectations rose from 11% to 19% in grade 6, 13% to 15% in grade 7 and 13% to 23% in grade 8 between 2021 and 2024, results of the combined program rather than MATHia alone.

Spokane Public Schools, a Washington district of 28,280 students across grades pre-K to 12, has used Carnegie Learning's high school math solution, including the adaptive tutor MATHia, since 2018. Carnegie Learning reports that for every 10 point increase in a student's MATHia Adaptive Personalized Learning Score, Spokane students saw a 14 to 28 point increase on the state Smarter Balanced Assessment (SBA) scale, and that every ten MATHia workspaces a student completed was associated with a 7 to 22 point increase on the same scale, with higher SBA pass rates among students who completed MATHia workspaces than among those who did not.

Muleshoe Independent School District, a Texas district, adopted the Texas Math Solution, which combines a TEKS aligned consumable text with MATHia, Carnegie Learning's adaptive one to one math coaching software, in its Geometry classes. Geometry teacher Xavier Mora had typically seen about 10% average growth between beginning and end of year assessment scores; in the 2021-22 school year, his first year using the Texas Math Solution and MATHia, his students showed an average 48% improvement.

Carnegie Learning's case study, which notes that Mora later joined the company as a Manager of School Partnerships, credits that growth to the Texas Math Solution as a whole (text and software together), not to MATHia alone, and says MATHia fosters student independence. Mora now has more Geometry teachers on his team using the programme.