aceModerate evidence
More
Meta-analysis of 35 studies finds AI helps students self-regulate, mostly during tasks
Study title: AI support in self‐regulated learning: A decade of technological evolution and meta‐analysis
- Evidence type
- Systematic review or meta-analysis
- Sample
- n = 35
- Sources
- 1 sourcePeer-reviewed
- Published
- 2 min read
- Truvace record
- TRV-2026-2816cite or check it
Study title: AI support in self‐regulated learning: A decade of technological evolution and meta‐analysis
A meta-analysis of 35 empirical studies published between 2013 and 2025 examined AI applications within Zimmerman's cyclical model of self-regulated learning. It found a moderate overall effect of g = 0.507 for AI-supported interventions.
Read the study abstract
Abstract This meta‐analysis systematically examines 35 empirical studies (2013–2025) investigating artificial intelligence applications within Zimmerman's cyclical model of self‐regulated learning (SRL). Three principal discoveries emerge: (1) Technological progression has evolved through three co‐existing paradigms: rule‐based architectures, data‐driven adaptive systems and generative AI ecosystems, that demonstrate increasingly sophisticated capabilities for human‐AI collaboration. (2) While AI‐supported SRL i…
62Gain side
How the two sides scoreClose call
62Problem side
The gain
A meta-analysis of 35 studies (2013–2025) found AI-supported self-regulated learning produced a moderate overall effect (g = 0.507), with generative AI strongest (g = 0.938 in the task performance phase).
The problem
A meta-analysis of 35 studies on AI-supported self-regulated learning suggests AI's ability to scaffold immediate task performance may outpace its ability to build durable, transferable self-regulatory skills.
A fixed formula sets both scores from the evidence and its sources. Reader votes don’t move them.
What the scores mean
G is the gain side, out of 100. It goes up when the gain is specific, when it measurably helped people and when the sources behind it are strong.
P is the problem side, out of 100. It goes up when the problem is specific, when it measurably hurt people and when the sources behind it are strong.
In brief
The effect varied by learning phase: g = 0.574 during task performance, g = 0.401 in forethought and g = 0.464 in self-reflection. Generative AI outperformed other approaches in every phase, though high heterogeneity suggests results depend heavily on instructional design.
The authors describe a performance-competence divide. AI's effect on observable behaviours (g = 0.751) was more than double its effect on self-reported perceptions (g = 0.369), and they say immediate task support may outpace lasting skill building.
Main points
- The analysis covered 35 empirical studies from 2013 to 2025 on AI within Zimmerman's cyclical model of self-regulated learning.
- Overall effect size was moderate (g = 0.507).
- AI worked better during task performance (g = 0.574) than in forethought (g = 0.401); self-reflection was g = 0.464.
- Generative AI showed the strongest effects, with g = 0.709 for forethought and g = 0.938 for performance.
- Effects were largest in secondary education, natural sciences, fully online settings and interventions lasting 2–10 weeks.
- Behavioural-trace measures showed g = 0.751 versus g = 0.369 for self-reports.
- Three AI paradigms coexist: rule-based, data-driven adaptive and generative AI.
The full story
The review traced a decade of technological change through three coexisting paradigms: rule-based architectures, data-driven adaptive systems and generative AI ecosystems. It says these show increasingly sophisticated capacity for human-AI collaboration.
Moderator analysis pointed to the best conditions: secondary education, natural science subjects, fully online delivery and interventions of medium length. The authors advise 2–10 weeks to avoid both novelty effects and scaffold dependency.
The authors recommend combining rule-based systems' stable scaffolding with generative AI's adaptive dialogue, and tailoring support by discipline, with social sciences needing more help for open-ended inquiry. They also call for assessments that measure delayed, unsupported performance alongside immediate gains.
What this doesn’t fix
The authors report high heterogeneity, suggesting the effects of generative AI are heavily contingent on specific instructional designs.
Sources
1 on file- British Journal of Educational Technologydoi:10.1111/bjet.70058
Talk it over
This record found a gain and a problem in the same place. Which matters more to you, and who might answer differently?
Meta-analysis of 35 studies finds AI helps students self-regulate, mostly during taskshttps://truvace.com/item/ai-support-in-self-regulated-learning-a-decade-of-technological-evolution-and-me
This record found a gain and a problem in the same place. Which matters more to you, and who might answer differently?
The gain A meta-analysis of 35 studies (2013–2025) found AI-supported self-regulated learning produced a moderate overall effect (g = 0.507), with generative AI strongest (g = 0.938 in the task performance phase).
The problem A meta-analysis of 35 studies on AI-supported self-regulated learning suggests AI's ability to scaffold immediate task performance may outpace its ability to build durable, transferable self-regulatory skills.
Answer in the threadHow do we know? What can this kind of evidence show, and what can’t it?
The evidence This rests on a systematic review or meta-analysis with a sample of 35, from British Journal of Educational Technology.
Answer in the threadWhat does it leave open? What else would you want to know before acting on it?
The record’s own caveat The authors report high heterogeneity, suggesting the effects of generative AI are heavily contingent on specific instructional designs.
Answer in the threadHave you seen AI used this way where you study, work or live? Did it go the same way?
Answer in the thread
The questions are the same on every record. The details under them are taken from this record as written.
What readers say