CoChat

Between cognitive offloading and critical autonomy: A systematic review of the epistemic implications of generative AI in higher education.

Open access
Suazo Galdames, Iván Claudio, Santiesteban Velázquez, Meylin, Saracostti, Mahia, Chaple Gil, Alain Manuel · 2026 · Frontiers in Education

The DOI matches its Crossref record; 48 of the 50 references it cites were found in a registry or an index; 19 of the 32 sources its claims cite were checked for support.

Abstract

Introduction: The rapid integration of generative artificial intelligence (GenAI) into higher education has generated increasing concern regarding its influence on students' epistemic agency, critical thinking, and cognitive autonomy. This systematic review synthesizes current evidence on the epistemic implications of GenAI in higher education, with particular emphasis on the tension between cognitive offloading and critical autonomy. Methods: Following PRISMA 2020 guidelines, peer-reviewed studies published between 2022 and 2026 were identified through systematic searches in Scopus, the Web of Science Core Collection, and seven education and multidisciplinary databases accessed through EBSCOhost. Eligible studies examined the educational use of generative AI systems such as ChatGPT, Gemini, Claude, and similar large language models within higher education contexts. Data extraction and thematic synthesis were conducted to identify recurring epistemic, cognitive, pedagogical, and ethical patterns across the included studies. Results: The findings reveal that GenAI simultaneously functions as a cognitive scaffold and a source of epistemic vulnerability. While AI-assisted learning environments may support feedback literacy, metacognitive reflection, conceptual clarification, and self-regulated learning, uncritical reliance on AI-generated outputs was frequently associated across the included studies with cognitive offloading, automation bias, superficial information processing, diminished analytical engagement, and weakened evaluative judgment. The evidence further suggests that the educational consequences of GenAI are mediated primarily by pedagogical design, AI literacy, and assessment practices rather than by the technology itself. Discussion: Overall, the review supports the need for higher education institutions to promote critical AI literacy and pedagogical frameworks capable of preserving epistemic agency, reflective reasoning, and cognitive autonomy in increasingly AI-mediated learning environments. Systematic Review Registration: https://www.crd.york.ac.uk/PROSPERO/view/CRD420261390452 , PROSPERO CRD420261390452.

Overview

CoChat Summary

TL;DR

This systematic review synthesizes 26 empirical studies on generative AI in higher education and finds a context-dependent tension: GenAI can act as a cognitive scaffold, but uncritical reliance is repeatedly associated with cognitive offloading, automation bias, superficial processing, and weakened evaluative judgment. The review proposes the HEEP-HE model, arguing that educational consequences depend mainly on pedagogical design, AI literacy, and assessment practices rather than on the technology itself.

Key findings

  • Across the included studies, GenAI functioned simultaneously as a cognitive scaffold and a source of epistemic vulnerability: AI-assisted learning was linked to feedback literacy, metacognitive reflection, conceptual clarification, and self-regulated learning, while uncritical reliance was associated with cognitive offloading, automation bias, superficial information processing, diminished analytical engagement, and weakened evaluative judgment.
  • The review included 26 studies: 3 published in 2024, 14 in 2025, and 9 in 2026; no eligible studies were published in 2022 or 2023. Geographically, 18 studies originated in the Global South, 7 in the Global North, and 1 had a mixed North–South perspective.
  • Methodologically, the corpus comprised 11 qualitative studies, 7 mixed-methods studies, 4 analytical cross-sectional studies, 2 quasi-experimental studies, and 2 randomized controlled trials. Inter-reviewer agreement was high, with an overall Cohen’s kappa of κ = 0.85.
  • The synthesis identified six descriptive themes and five analytical themes, including reconfiguration of epistemic authority, fragility of knowledge validation, transformation of academic integrity into epistemic integrity, cognitive offloading and erosion of learner agency, GenAI as scaffold for higher-order thinking, and institutional transition from prohibition to governed integration. The HEEP-HE model organizes these into epistemological validation, pedagogical scaffolding, sociotechnical calibration, and civic-democratic resilience.

Methods

  • Systematic review with thematic and framework synthesis, reported according to PRISMA 2020 and prospectively registered in PROSPERO (CRD420261390452). Searches were conducted on May 2, 2026, in Scopus, Web of Science Core Collection, and seven education/multidisciplinary databases via EBSCOhost.
  • Eligible studies were peer-reviewed empirical articles published between January 2022 and May 2026, situated in higher education, and addressing epistemic dimensions of GenAI use such as critical thinking, epistemic trust, source evaluation, cognitive offloading, AI literacy, academic integrity, or metacognitive engagement. Grey literature, reviews, conceptual articles, and purely technical studies were excluded.
  • Two independent reviewers screened, extracted data, and appraised quality using JBI critical appraisal tools. Thematic synthesis followed Thomas and Harden’s three-stage procedure, and framework synthesis mapped evidence onto four domains. Narrative synthesis followed SWiM; meta-analysis was not considered appropriate due to heterogeneity.

Limitations

  • The paper states that the search was restricted to Scopus, Web of Science Core Collection, and EBSCOhost databases, so relevant studies indexed elsewhere may have been missed; grey literature was excluded, potentially increasing publication and availability bias; and the search ended on May 2, 2026, requiring relatively early updating.
  • The paper also states that inclusion of early-access/online-first publications means bibliographic details may change; conceptual and methodological heterogeneity precluded statistical pooling; much evidence was cross-sectional, perception-based, or self-reported, restricting causal inference and introducing possible recall and social-desirability bias; and included studies operationalized critical autonomy, cognitive offloading, epistemic agency, and AI literacy inconsistently.
  • The paper acknowledges that thematic and framework synthesis involve interpretive judgment, and that the HEEP-HE model is a provisional evidence-informed framework requiring independent empirical evaluation rather than a validated intervention or causal model.

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

Integrity signals

Registry

Crossref · 10.3389/feduc.2026.1878665

Cites

48 of 50 references verified0 citation errors · 2 not checkable

Statistics

Nothing to recompute.

Figures

Image 1
CC-BY · From the PDF
Image 2
CC-BY · From the PDF

Verification

Claims the paper makes

18 Valid · 1 Not valid · 3 Not checked · 192 Own results and interpretations

79 more claims

Create a free account to see all 214 claims and every source verdict.

See all claims — free account
  • Probably not a substantial claim. 94
  • It may state two separate claims as one. 55

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Citations

Cites

Research works printed but not linked: 2

  1. Fujimoto, 2026, A literature review on the use of generative ai in English language education, ICIC Express Lett. B: Appl., 17, 391, 10.24507/icicelb.17.04.391 · doi:10.24507/icicelb.17.04.391
  2. Thi, 2024, Understanding Vietnamese English majors’ use and acceptance of AI-powered tools for English academic writing at university, Taiwan Int. ESP J., 15, 29, 10.6706/TIESPJ.202412_15(2).0002 · doi:10.6706/tiespj.202412_15(2).0002

Cited by

See all 5

Metadata

Authors
Suazo Galdames, I. C., Santiesteban Velázquez, M., Saracostti, M., & Chaple Gil, A. M.
Affiliations
  • Vicerrectoría de Investigación y Doctorados, Universidad Autónoma de Chile
  • Facultad de Odontología, Universidad Autónoma del Paraguay
  • Departamento de Trabajo Social, Universidad de Chile
  • Facultad de Ciencias de la Salud, Carrera de Odontología, Universidad Autónoma de Chile
Venue
Frontiers in Education
Published
2026-09-03
Volume
11
ISSN
2504-284X (ISSN-L)
OpenAlex
W7207563094
Cited by
5
References
41
Citations by year
2026: 5
FWCI
8.97
Type
journal article
Language
en
Open access
Locations
  • Frontiers in Education (publishedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (acceptedVersion)
Topics
  • Social Sciences › Education › Artificial Intelligence in Education
  • Social Sciences › Psychology › Developmental and Educational Psychology › Educational Strategies and Epistemologies
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
Keywords
generative AI, higher education, cognitive offloading, critical autonomy, epistemic agency, critical thinking, large language models, AI literacy, automation bias, student self-regulation, metacognition, assessment practices, instructional design, AI-mediated learning

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 2026-10-20, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-20 (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

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.