Learning research, in plain words
What new studies on learning actually found, for parents, homeschoolers, teachers and anyone teaching themselves. Each summary covers what the researchers did, who took part, what they found, effect sizes where reported, and the limitations.
15 summaries 路 last updated 路 see all 路 how we write these
Newest
- With AI, people scored better on reasoning puzzles but overestimated how well they did 路 Daniela Fernandes et al., 2026 路 Two online experiments
- Unrestricted ChatGPT help raised practice scores but lowered exam scores 路 Hamsa Bastani et al., 2025 路 Preregistered randomized controlled trial
- Six weeks of teacher-guided AI tutoring improved English results in Nigeria 路 Martin De Simone et al., 2025 路 Randomized controlled trial
- A carefully designed AI tutor beat an active-learning physics class on a short-term test 路 Greg Kestin et al., 2025 路 Randomized controlled trial with a crossover design
- "Your Brain on ChatGPT": what the MIT essay-writing study did and didn't show 路 Nataliya Kosmyna et al., 2025 路 Lab experiment with EEG
- Supervised AI tutoring matched human tutors in UK maths classrooms 路 LearnLM Team (Google) et al., 2025 路 Exploratory randomized controlled trial
By topic
AI tutors and learning outcomes
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.
- Unrestricted ChatGPT help raised practice scores but lowered exam scores 路 Hamsa Bastani et al., 2025 路 Preregistered randomized controlled trial
- A carefully designed AI tutor beat an active-learning physics class on a short-term test 路 Greg Kestin et al., 2025 路 Randomized controlled trial with a crossover design
- Supervised AI tutoring matched human tutors in UK maths classrooms 路 LearnLM Team (Google) et al., 2025 路 Exploratory randomized controlled trial
- Six weeks of teacher-guided AI tutoring improved English results in Nigeria 路 Martin De Simone et al., 2025 路 Randomized controlled trial
- An AI coach for human tutors raised topic mastery, most for weaker tutors 路 Rose Wang et al., 2024 路 Preregistered randomized controlled trial
"AI brain rot" and cognitive offloading
Studies on what happens to effort, reasoning and memory when an AI does part of the thinking. Lower effort is not always a problem, but these studies show when it costs learning.
- With AI, people scored better on reasoning puzzles but overestimated how well they did 路 Daniela Fernandes et al., 2026 路 Two online experiments
- Unrestricted ChatGPT help raised practice scores but lowered exam scores 路 Hamsa Bastani et al., 2025 路 Preregistered randomized controlled trial
- "Your Brain on ChatGPT": what the MIT essay-writing study did and didn't show 路 Nataliya Kosmyna et al., 2025 路 Lab experiment with EEG
- ChatGPT made research feel easier but students' reasoning was weaker than with Google 路 Matthias Stadler et al., 2024 路 Randomized experiment
Socratic and dialogic tutoring
Research on tutoring that asks guiding questions and gives hints instead of handing over solutions, including AI that drafts those questions.
- Unrestricted ChatGPT help raised practice scores but lowered exam scores 路 Hamsa Bastani et al., 2025 路 Preregistered randomized controlled trial
- Supervised AI tutoring matched human tutors in UK maths classrooms 路 LearnLM Team (Google) et al., 2025 路 Exploratory randomized controlled trial
- An AI coach for human tutors raised topic mastery, most for weaker tutors 路 Rose Wang et al., 2024 路 Preregistered randomized controlled trial
- Prompting learners to explain to themselves helps: a meta-analysis of 64 reports 路 Kiran Bisra et al., 2018 路 Meta-analysis
Metacognition: knowing what you know
How well learners judge their own understanding, and why some study methods and AI tools make that judgement worse.
- With AI, people scored better on reasoning puzzles but overestimated how well they did 路 Daniela Fernandes et al., 2026 路 Two online experiments
- "Your Brain on ChatGPT": what the MIT essay-writing study did and didn't show 路 Nataliya Kosmyna et al., 2025 路 Lab experiment with EEG
- Prompting learners to explain to themselves helps: a meta-analysis of 64 reports 路 Kiran Bisra et al., 2018 路 Meta-analysis
- The classic study: testing yourself beats rereading after a week 路 Henry L. Roediger III et al., 2006 路 Lab experiments
Retrieval practice (the testing effect)
Pulling information out of memory, by quizzing, blurting or flashcards, is one of the best-replicated ways to make learning last.
- Quizzing boosts classroom learning: a meta-analysis of 222 studies 路 Chunliang Yang et al., 2021 路 Systematic review and meta-analysis
- The classic study: testing yourself beats rereading after a week 路 Henry L. Roediger III et al., 2006 路 Lab experiments
Spaced repetition
Spreading study over days beats cramming the same amount into one sitting. The question is how far apart.
- Spacing beats cramming: the classic review of 317 experiments 路 Nicholas J. Cepeda et al., 2006 路 Review and meta-analysis
Self-explanation
Explaining to yourself why each step works, while you study or solve problems.
- Prompting learners to explain to themselves helps: a meta-analysis of 64 reports 路 Kiran Bisra et al., 2018 路 Meta-analysis
Homeschooling outcomes
What longitudinal data say about how homeschooled young people fare, and how careful we have to be with these associations.
- Homeschooled teens in a large US cohort: more volunteering, fewer college degrees 路 Ying Chen et al., 2021 路 Longitudinal cohort study
Voice and learning out loud
Reading aloud, explaining out loud and talking things through: what the evidence does and does not show.
- Reading aloud helps you remember a text, but not understand it better 路 Brady R. T. Roberts et al., 2023 路 Four lab experiments
Motivation
Which kinds of motivation go with learning, persistence and well-being, according to large syntheses.
- A carefully designed AI tutor beat an active-learning physics class on a short-term test 路 Greg Kestin et al., 2025 路 Randomized controlled trial with a crossover design
- Which kinds of motivation go with doing well: a meta-analysis of 223,209 students 路 Joshua L. Howard et al., 2021 路 Meta-analysis
Made by the Ask Poppy team, who also make free study tools at learn.askpoppy.ai and an AI tutor app. We say so wherever it comes up.