AI tutors and learning outcomes: research in plain words

What controlled studies say so far about AI tutors: when they help, when they hurt, and why the design of the tutor matters more than the model.

Summaries (5)

Unrestricted ChatGPT help raised practice scores but lowered exam scores

Hamsa Bastani et al. (2025) · Preregistered randomized controlled trial (field experiment) · Peer-reviewed

In a high school in Turkey, students who practised maths with a plain ChatGPT-style GPT-4 assistant did much better during practice, then scored 17% lower than students without AI when they sat a test on their own. A second version with teacher-written prompts that gave hints instead of solutions avoided that drop, but did not beat the no-AI group on the test either.

A carefully designed AI tutor beat an active-learning physics class on a short-term test

Greg Kestin et al. (2025) · Randomized controlled trial with a crossover design · Peer-reviewed

In a Harvard introductory physics course, students learned more on a short post-test when a lesson was delivered by a custom-built AI tutor at home than when the same material was taught as an in-class active-learning lesson. The tutor was not a plain chatbot: instructors wrote question-specific prompts and solutions and structured each step.

Supervised AI tutoring matched human tutors in UK maths classrooms

LearnLM Team (Google) et al. (2025) · Exploratory randomized controlled trial · Preprint (not peer-reviewed at the time of summary)

In five UK secondary schools, pupils who got chat tutoring drafted by Google's LearnLM model, with every message checked by an expert tutor, did at least as well as pupils tutored by humans alone, and were 5.5 percentage points more likely to solve new problems on later topics. Tutors approved most AI drafts with little or no editing.

Six weeks of teacher-guided AI tutoring improved English results in Nigeria

Martin De Simone et al. (2025) · Randomized controlled trial · Working paper

Senior secondary students in Benin City, Nigeria, who attended twelve after-school English sessions using Microsoft Copilot, with teachers guiding and suggesting prompts, scored about 0.3 standard deviations higher on a final assessment than students who did not. They also did better on the regular end-of-term English exam, which covered more than the programme did.

An AI coach for human tutors raised topic mastery, most for weaker tutors

Rose Wang et al. (2024) · Preregistered randomized controlled trial · Working paper / preprint

Giving human tutors an AI tool that suggests expert-style next moves made their students 4 percentage points more likely to master a topic, and 9 points more likely for students of lower-rated tutors. Tutors with the tool asked more guiding questions and gave away answers less often. End-of-year test scores did not improve significantly.

Free tools for this

Each summary says what the researchers did, who took part, what they found, the limitations, and whether we read the full paper or only the abstract.