Truace
Subscribe
70/100

Moderate evidence

Cite
More
Export CSVExport JSONEmailLinkedInX

Review of 32 studies shows AI tools helped learners plan and direct their own study

Study title: Artificial intelligence and learner autonomy: a meta-analysis of self-regulated and self-directed learning

Evidence type
Systematic review or meta-analysis
Sample
n = 3,029
Sources
1 sourcePeer-reviewed
Published
2 min read
Truvace record
TRV-2026-2823cite or check it

Study title: Artificial intelligence and learner autonomy: a meta-analysis of self-regulated and self-directed learning

A meta-analysis in Frontiers in Education pooled 32 studies that compared AI-supported learners with control groups, covering 92 effect sizes and 3,029 participants. AI-based interventions had a large positive effect on overall self-regulated learning (g = 1.613) and on self-directed learning (g = 1.111), and a smaller effect on learning outcomes and achievement (g = 0.350).

Read the study abstract

Introduction As artificial intelligence (AI) becomes increasingly embedded in educational environments, understanding its role in shaping learners’ self-regulated learning (SRL) and self-directed learning (SDL) has emerged as a central concern in contemporary learning science. While prior studies suggest that AI-driven systems may support planning, monitoring, and autonomy in learning, empirical evidence remains fragmented across contexts, learner groups, and instructional designs. This study synthesizes existin…

The gain

A meta-analysis of 32 controlled studies with 3,029 learners found AI-based interventions raised self-regulated learning (g = 1.613) and self-directed learning (g = 1.111) versus control groups, with very high variation between studies.

In brief

The gains were uneven. AI helped most with planning and goal setting before a task and gave smaller gains in self-reflection afterwards, but effects on behavioural regulation such as time management and effort were inconsistent and not statistically significant. Effects were stronger among older learners, in longer interventions and in language learning with interactive AI systems.

The large headline figures come with very high variation between studies (I2 of 98.85% for overall self-regulated learning). Tests also flagged possible publication bias for the overall result, so the size of the effect is less certain than its positive direction.

Main points

  1. The review searched Scopus and Google Scholar on 27 September 2025, found 427 documents and kept 32 studies with a control group; participants were aged 18 to 37, mostly undergraduates and early postgraduates.
  2. Overall self-regulated learning rose by g = 1.613 (p = 0.032; 11 studies; 95% CI 0.14 to 3.09), and self-directed learning rose by g = 1.111 (p = 0.043; 8 studies).
  3. Cognitive and metacognitive regulation improved by g = 0.377 (p = 0.0004; 10 studies) and motivational and affective regulation by g = 0.505 (p = 0.013); behavioural regulation showed g = 0.914 but was not significant (p = 0.284).
  4. By phase, the forethought phase gained g = 0.4739 and self-reflection g = 0.2774; the performance phase (g = 0.78) had a confidence interval of -0.11 to 1.67 and was not significant.
  5. Learning outcomes and achievement improved by g = 0.350 (p = 0.034) across seven studies, a small-to-moderate effect.
  6. Older learners and longer interventions showed larger gains in overall self-regulation; gender made no significant difference.
  7. Conversational and interactive systems such as chatbots produced strong effects on nearly all outcomes, while instructional and adaptive systems showed no significant moderating effect.

What the review found when it combined the studies

These are the review’s own pooled results, copied from its abstract. Truvace did not recalculate them.

  • Learners' self-regulated learning during the performance phase with AI-based interventions

    Hedges' g 0.78 (95% CI −0.11 to 1.67)

    32 studies · 3,029 people

    The review’s own words
    Although the performance phase (Figure 9) yielded a large mean effect size (g = 0.78), the confidence interval includes zero (−0.11 to 1.67)

Code checks every number above against the review’s text before it is shown. The short labels are an AI model’s reading of that text, so the review’s own sentence sits under each one. A pooled figure is only as strong as the studies behind it. How Truvace uses pooled results

The full story

The studies covered language learning, STEM and medical education, using tools ranging from chatbots and a Study Buddy SRL chatbot to LLaMA-based AI textbooks, analytics-enhanced feedback systems and rule-based scaffolding engines. Effect sizes were Hedges' g, pooled with random-effects models and robust variance estimation where studies reported several effects.

Heterogeneity differed sharply by outcome. Cognitive and metacognitive regulation had low variation (I2 = 30.98%) and forethought effects moderate variation (about 44%), while behavioural regulation (98.32%) and the performance phase (about 97%) varied widely across studies. The author reads this as AI supporting in-the-head planning and monitoring better than habits that depend on a learner's environment.

Leave-one-out checks kept the overall self-regulation effect between g = 0.914 and 1.746, so no single study drove it. Egger's and Begg's tests nonetheless indicated significant publication bias for overall self-regulated learning, and Begg's test also flagged self-directed learning (p = 0.0098).

The author cautions that opaque or overly prescriptive AI recommendations could make learners follow generated paths rather than direct their own study, and that some learners may offload effort instead of building it.

Sources

1 on file
  1. Frontiers in Educationdoi:10.3389/feduc.2025.1738751
    Peer-reviewed2025-12-29

Talk it over

Four questions to start onWho gets this benefit, and who was left out of the study or story?

Start with a question

Review of 32 studies shows AI tools helped learners plan and direct their own studyhttps://truvace.com/item/artificial-intelligence-and-learner-autonomy-a-meta-analysis-of-self-regulated-a

  1. Who gets this benefit, and who was left out of the study or story?

    The gain A meta-analysis of 32 controlled studies with 3,029 learners found AI-based interventions raised self-regulated learning (g = 1.613) and self-directed learning (g = 1.111) versus control groups, with very high variation between studies.

    Answer in the thread

  2. How 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 3,029, from Frontiers in Education.

    Answer in the thread

  3. What does it leave open? What else would you want to know before acting on it?

    The record’s own caveat Variation between studies was extremely high for overall self-regulated learning, and tests indicated significant publication bias for that result, so the pooled effect size may be overestimated.

    Answer in the thread

  4. Have 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

More from Policy