The exact text we saved

This is the version exactly as it was saved, character for character. The SHA-256 hash below is a 64-character code worked out from that text. Run the same text through SHA-256 and you get the same code; change a single letter and the code comes out completely different. Each version also stores the previous version’s hash, which chains them together so no version can be swapped out quietly.

TRUVACE RECORD VERSION
record: TRV-2026-2721
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-11-12T17:14:40.878643Z
status: published
lens: g_space
sector: policy
headline: A meta-analysis of AI-assisted self-regulated learning
dek: This meta-analysis examines the effects of artificial intelligence (AI)-based pedagogical interventions on learners’ self-regulated learning and its subdomains, including cognitive, metacognitive, and resource management strategies. In addition, the study investigates whether these effects vary across four groups of moderators concerning research context, intervention design, AI, and self-regulated learning. A systematic search of databases identified an initial pool of 5,741 potentially relevant studies, of whi…
gain_title: A meta-analysis of 34 experimental and quasi-experimental studies (91 effect sizes) found AI-based teaching interventions improved students' self-regulated learning strategies by Hedges' g = 0.76 compared with control groups.
problem_title: (none)
trace_subject: (none)
gain_reading: A meta-analysis of 34 experimental and quasi-experimental studies (91 effect sizes) found AI-based teaching interventions improved students' self-regulated learning strategies by Hedges' g = 0.76 compared with control groups.
gain_evidence: Results showed a positive overall effect of AI-based pedagogical interventions on self-regulated learning (Hedges’ g = 0.76).
problem_reading: (none)
problem_evidence: (none)
quick_read: A meta-analysis led by Zhuo Zhang of Towson University pooled 34 studies, drawn from 5,741 records, that compared AI-based teaching interventions with control groups. Across 91 effect sizes, the interventions raised students' self-regulated learning by Hedges' g = 0.76, which the authors call a moderate-to-large effect (95% CI 0.52 to 1.01).

The size of the gain differed by strategy type. Resource management strategies such as help-seeking and time management improved most (g = 0.93), followed by metacognitive strategies such as planning and self-evaluation (g = 0.73) and cognitive strategies such as rehearsal and elaboration (g = 0.48).

Results varied widely between studies (I² = 88.34%), and the evidence leans heavily on university students: 25 of 34 studies were in higher education, 5 in secondary and 4 in elementary schools. The authors say the true effects in K-12 settings remain uncertain.
limitation: Results varied widely between studies (I² = 88.34%), higher education dominated the evidence base (25 of 34 studies), and the authors treat the moderator findings as exploratory and tentative.
tag: Evidence-backed gain
key_points: The search covered ERIC, Education Source, PsycINFO, Web of Science, Scopus and ProQuest and identified 5,741 records; 216 reports went to full-text review and 34 met all criteria. | Overall effect on self-regulated learning was g = 0.76 (SE 0.12, p < .001) across 91 effect sizes. | Subdomain effects: resource management g = 0.93 (11 studies), metacognitive g = 0.73 (20 studies), cognitive g = 0.48 (9 studies). | Intelligent tutoring or adaptive learning systems showed the largest mean effect (g = 1.20), followed by conversational agents (0.88), LLM-supported systems (0.56) and learning analytics dashboards (0.37). | AI roles giving feedback or scaffolding beat monitoring-only tools (g = 0.32); scaffolding or facilitation was significantly stronger than monitoring or analytics. | Generative and non-generative AI did not differ significantly (g = 0.59 versus 0.88, p > .05). | Education level mattered: higher education g = 0.83, elementary g = 0.81, secondary g = 0.41, though the elementary-secondary gap was borderline (p = .04996).
rundown: The authors included only studies with experimental or quasi-experimental designs and a passive or active control group, and they excluded one-group pre-post designs. The 34 reports were 30 journal articles and four conference papers; 15 were experimental and 19 quasi-experimental. Most were in language learning (13) and science (9), and 15 lasted less than a week.

Study quality was modest. Appraisal scores on a six-point scale ranged from 1 to 4 (mean 2.76); three articles scored 1 and six scored 4. No study was excluded on quality grounds. Effects were calculated with robust variance estimation to handle multiple effect sizes per study.

Moderator results were mixed. Discipline, intervention length, control group type, sample size and how self-regulated learning was measured showed no significant differences. Studies with small to medium samples reported larger effects than those with more than 150 participants.

The authors say publication bias is unlikely to change the findings: Egger's test was not significant and trim-and-fill found no missing studies. They add that the funnel plot showed some asymmetry, so small-study effects cannot be ruled out.
sources:
- peer_reviewed | Online Learning | https://doi.org/10.24059/olj.v30isi2.5691 | 2026-11-06
prev: 0000000000000000000000000000000000000000000000000000000000000000

SHA-256 hash

d78c5a7f9d26e7618a9c85882853460eb05841eabebd9398c548e63fc0ed34ac

Previous version’s hash

0000000000000000000000000000000000000000000000000000000000000000