CoChat

Enhancing School Students' Self-Regulated Learning through Generative AI Support: A Randomized Controlled Trial

Open access
Fütterer, Tim, Bardach, Lisa, Kuhn, Jochen, Keller, Stefan Daniel, Gerjets, Peter · 2026 · Educational Psychology Review

The DOI matches its Crossref record; 138 of the 139 references it cites were found in a registry or an index; 38 of the 74 sources its claims cite were checked for support.

Abstract

Abstract Self-regulated learning (SRL) is a critical competency for academic success and lifelong learning, yet providing personalized SRL support in real-time classroom settings remains challenging. In this randomized controlled trial, we investigated whether generative-AI-(GenAI)-based interventions can effectively support SRL in secondary school students. Drawing on an integrative model of self-regulation in education, the study contrasted two types of GenAI-supported interventions—targeting either motivational (utility value) or strategic (cognitive-learning-strategy-based) aspects of SRL—with a control condition using standard ChatGPT. A total of 371 students (Grades 7–9) were randomly assigned to one of the three conditions and participated in six 45-min sessions during regular physics or English lessons. Outcome measures included self-reported utility value, strategy use (self-reported and tested performance), interest, effort, and tested domain-specific knowledge, assessed at pre- and posttest. The results showed that students in the utility value condition reported more favorable development of perceived utility value than those in the cognitive strategy condition. However, no statistically significant advantages of either intervention over the control condition were found for effort, domain-specific knowledge, or elaboration-based strategy use. Exploratory analyses indicated that students who engaged more meaningfully with the GenAI tended to have more sustained interest than those in the control group. No subject-specific (physics or English) differences in intervention effects were observed. Findings suggest GenAI-based interventions may help preserve motivational aspects of SRL under certain conditions, but further development is needed to effectively support cognitive strategies and improve learning outcomes in secondary school settings.

Overview

CoChat Summary

TL;DR

This preregistered RCT tested whether two types of GenAI-supported prompting—utility value reflection and cognitive learning strategy prompting—could foster self-regulated learning in German secondary school students, compared with standard ChatGPT use. Only the utility value condition showed a relative advantage, and only over the cognitive strategy condition (not the control); no condition effects emerged for strategy use, effort, or domain-specific knowledge.

Key findings

  • Utility value: the time × condition interaction was significant, χ²(2) = 19.05, p < .001. Students in the utility value condition showed a more favorable trajectory than those in the cognitive learning strategy condition (β = −0.50, SE = 0.11, p < .001), driven by a significant decline in the cognitive strategy condition (ΔM = 0.30, p < .001) rather than a gain in the utility value condition (ΔM = −0.08, p = .328). The utility value condition did not differ significantly from the control condition (ΔM = −0.15, p = .063).
  • Strategy use: no differential change across conditions for self-reported elaboration-based strategy use, χ²(2) = 0.08, p = .960, or for applied learning-by-explaining, χ²(2) = 0.01, p = .992; all conditions showed slight descriptive decreases (ΔM ≈ 0.32–0.35) that were not significant.
  • Other outcomes: no condition × time effects for maintained interest (χ²(1) = 0.54, p = .463), effort investment (χ²(1) = 0.24, p = .623), or domain-specific knowledge (χ²(1) = 1.28, p = .257).
  • Exploratory: a significant three-way interaction of condition, time, and relative number of meaningful GPT interactions for maintained interest (β = 0.30, SE = 0.00, p = .009), indicating that meaningful engagement buffered motivational decline in the intervention conditions relative to control. No subject-specific (physics vs. English) differences emerged for any outcome.

Methods

  • Preregistered three-arm RCT in authentic classrooms (physics and English) in Baden-Württemberg, Germany; N = 371 students in Grades 7–9 (age M = 13.92, SD = 1.27; 54% female; 45% reported a migration background), individually randomized within the IWM Author platform to utility value reflection, cognitive learning strategy prompting, or a standard-ChatGPT control condition.
  • Intervention spanned six 45-min sessions (pretest, four learning sessions, posttest) with GPT-4o-powered CustomGPTs; all conditions withheld correct answers, and conditions differed only in theory-informed system prompts. Outcomes at T1 and T2 included self-reported utility value, elaboration-based strategy use, applied learning-by-explaining, maintained interest, effort, and tested domain-specific knowledge.
  • Analysis used linear mixed-effects models (lme4) with condition, time, and their interaction, random intercepts for participants, emmeans post hoc contrasts, and Benjamini–Hochberg correction; robustness checks included multiple imputation (50 datasets), complete-case analysis, covariate adjustment per WWC standards, and CACE estimation via instrumental variable regression.

Limitations

  • Stated by the authors: convenience sampling limits generalizability beyond German secondary schools; unequal group sizes with no blocking or stratification may have introduced baseline imbalances; 35% dropout between T1 and T2 (130 students) risked attrition bias; several exploratory analyses increased false-positive risk; teacher presence may have introduced implementation variability; interaction measures captured engagement/compliance rather than process-level SRL behaviors; all outcomes were measured immediately post-intervention with no follow-up; and reliability of some self-report scales was low (e.g., openness α = 0.39; conscientiousness α = 0.59; AI literacy KR-20 = 0.23).
  • Evident from the text: the self-report elaboration measure captured dispositional rather than task-specific strategy use, which the authors note was misaligned with the intervention and hypothesis; the control condition was an active ChatGPT condition rather than a waitlist, which the authors identify as a possible reason for null effects; and the four-session intervention may have been too brief to change deep-level competencies such as English grammar structures.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1007/s10648-026-10133-8

Cites

138 of 139 references verified1 citation errors · 0 not checkable

Statistics

Figures

Image 1
From the PDF
Image 2
From the PDF
Image 3
From the PDF
Image 4
From the PDF
Image 5
From the PDF
Table 6 (continued)
Table 6 (continued)From the PDF
Image 7
From the PDF
Fig. 6 Estimated marginal means of self-reported use of the elaboration-based cognitive learning strategy by intervention condition and time. Note . Error bars represent 95% confidence intervals. uv = utility value condition, cl = cognitive learning condition, co = control condition
Fig. 6 Estimated marginal means of self-reported use of the elaboration-based cognitive learning strategy by intervention condition and time. Note . Error bars represent 95% confidence intervals. uv = utility value condition, cl = cognitive learning condition, co = control conditionFrom the PDF
Fig. 7 Estimated marginal means of the application of learning-by-explaining by intervention condition and time. Note . Error bars represent 95% confidence intervals. uv = utility value condition, cl = cognitive learning condition, co = control condition
Fig. 7 Estimated marginal means of the application of learning-by-explaining by intervention condition and time. Note . Error bars represent 95% confidence intervals. uv = utility value condition, cl = cognitive learning condition, co = control conditionFrom the PDF
Image 10
From the PDF
Image 11
From the PDF
Image 12
From the PDF
Image 13
From the PDF
Image 14
From the PDF
Image 15
From the PDF
Image 16
From the PDF
Image 17
From the PDF
Image 18
From the PDF
Image 19
From the PDF
Image 20
From the PDF
Image 21
From the PDF
Fig. 14 Interaction of condition and meaningful interactions with the GPTs regarding utility value. Note . uv = utility value condition, cl = cognitive learning condition, co = control condition
Fig. 14 Interaction of condition and meaningful interactions with the GPTs regarding utility value. Note . uv = utility value condition, cl = cognitive learning condition, co = control conditionFrom the PDF
Fig. 15 Interaction of condition and meaningful interactions with the GPTs regarding self-reported elaboration-based strategy use. Note . uv = utility value condition, cl = cognitive learning condition, co = control condition
Fig. 15 Interaction of condition and meaningful interactions with the GPTs regarding self-reported elaboration-based strategy use. Note . uv = utility value condition, cl = cognitive learning condition, co = control conditionFrom the PDF
Image 24
From the PDF
Image 25
From the PDF
Image 26
From the PDF
Image 27
From the PDF

Verification

Claims the paper makes

27 Valid · 5 Not valid · 24 Not checked · 173 Own results and interpretations

101 more claims · 5 with their cited sources checked

Create a free account to see all 229 claims and every source verdict.

See all claims — free account
  • Probably not a substantial claim. 85
  • It may state two separate claims as one. 55
  • The quote only partly states this claim. 3

Create a free account to see these claims.

Statistics recomputed

Every reported test result, recomputed from its test statistic and degrees of freedom. A result shows as not adding up only once an independent check has confirmed it.

  • Not confirmedχ²(2) = 0.01reported p = .992 · recomputed p = .995

    “χ2(2) = 0.01, p = .992”

  • Consistentχ²(2) = 19.05reported p < .001 · recomputed p < .001

    “χ2(2) = 19.05, p < .001”

  • Consistentχ²(2) = 0.08reported p = .96 · recomputed p = .961

    “χ2(2) = 0.08, p = .960”

  • Consistentχ²(1) = 0.54reported p = .463 · recomputed p = .462

    “χ2(1) = 0.54, p = .463”

  • Consistentχ²(1) = 0.24reported p = .623 · recomputed p = .624

    “χ2(1) = 0.24, p = .623”

5 of 9 · the rest aren’t loaded

4 more recomputed tests

Create a free account to see every test the Stats check recomputed.

See all tests — free account

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

1 of 1

Citations

Cites

Show all 139 references

Research works printed but not linked: 7

  1. Angrist, J. D., & Pischke, J. S. (2015). Mastering ’metrics. The path from cause to effect. Princeton University Press.
  2. Chi, M. T. H. (2000). Self-explaining: The dual processes of generating inference and repairing mental models. In R. Glaser (Ed.), Advances in instructional psychology: Educational design and cognitive science (Vol. 5, pp. 161–238). Lawrence Erlbaum Associates.
  3. Foster, N., & Piacentini, M. (2023). Innovating assessments to measure and support complex skills (1st ed.). Organization for Economic Cooperation & Development (OECD).
  4. Hausmann, R. G. M., & Vanlehn, K. (2007). Explaining self-explaining: A contrast between content and generation. Proceedings of the 2007 Conference on Artificial Intelligence in Education: Building Technology Rich Learning Contexts That Work, 417–424.
  5. John, O. P., Naumann, L. P., & Soto, C. J. (2008). Paradigm shift to the integrative Big Five trait taxonomy: History, measurement, and conceptual issues. In O. P. John, R. W. Robins, & L. A. Pervin (Eds.), Handbook of personality: Theory and research (3rd ed., pp. 114–158). The Guilford Press.
  6. Lightbown, P., & Spada, N. M. (2013). How languages are learned (4th ed.). Oxford University Press.
  7. Schunk, D. H., & Zimmerman, B. J. (Eds.). (1998). Self-regulated learning: From teaching to self-reflective practice. Guilford Press.

Cited by

See all 17

Metadata

Authors
Fütterer, T., Bardach, L., Kuhn, J., Keller, S. D., & Gerjets, P.
Affiliations
  • University of Tübingen (DE)
  • Kaunas University of Technology (LT)
  • Zurich University of Teacher Education (CH)
  • Leibniz-Institut für Wissensmedien (DE)
Corresponding author
Fütterer, Tim
Venue
Educational Psychology Review
Published
2026-04-14
Volume
38
Issue
1
ISSN
  • 1040-726X (ISSN-L)
  • 1573-336X
OpenAlex
W7154244448
Cited by
17
References
124
Citations by year
2026: 17
FWCI
245.63
Type
journal article
Language
en
Open access
Licence
Locations
  • Educational Psychology Review (publishedVersion, cc-by)
  • KTUePubl (Repository of Kaunas University of Technology) (submittedVersion, other-oa)
  • Universitätsbibliothek Tübingen
Funders
Eberhard Karls Universität Tübingen
Topics
  • Social Sciences › Psychology › Developmental and Educational Psychology › Innovative Teaching and Learning Methods
  • Social Sciences › Education › Artificial Intelligence in Education
  • Physical Sciences › Computer Science › Artificial Intelligence › Intelligent Tutoring Systems and Adaptive Learning
Keywords
student self-regulation, generative AI, ChatGPT, high school students, utility value, cognitive strategies, elaboration, student motivation, interest, effort, domain knowledge, AI-based interventions, physics education, English education

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-09-28, under version 3 of the claim definition (prompt extract-v3).

A second model labelled each claim on 2026-09-28 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Enhancing School Students' Self-Regulated Learning through Generative AI Support: A Randomized Controlled Trial. https://app.cochat.ai/papers/10.1007/s10648-026-10133-8
The paper
Fütterer, T., Bardach, L., Kuhn, J., Keller, S. D., & Gerjets, P. (2026). Enhancing School Students' Self-Regulated Learning through Generative AI Support: A Randomized Controlled Trial. Educational Psychology Review. https://doi.org/10.1007/s10648-026-10133-8

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.