Between cognitive offloading and critical autonomy: A systematic review of the epistemic implications of generative AI in higher education.
Open accessThe 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
Cites
Statistics
Verification
Claims the paper makes
18 Valid · 1 Not valid · 3 Not checked · 192 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Recent evidence suggests that these systems are no longer perceived as occasional technological aids, but rather as normalized components of academic learning ecosystems integrated into writing, information retrieval, summarization, brainstorming, and assessment preparation processes (Cooper et al., 2025;Susena et al., 2026).”
§ Between cognitive offloading and critical autonomy: a systematic review of the epistemic implications of generative AI in higher education > 1 Introduction
What the sources say
- Partly supportedcites Cooper et al., 2025 sourceExperimental — accuracy not yet measured
The Introduction (page not supplied) reports that pre-service teachers use GenAI for lesson ideas, summarisation and assessment preparation, and describes ChatGPT being used for queries. The section on implications for science education also discusses support for writing. But the paper does not establish that GenAI is a normalized component of academic learning ecosystems generally; it says integration remains uneven, particularly among in-service teachers. (LLM, conf=0.86)
- Not checked — no full text of the sourcecites Susena et al., 2026 source
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Recent scholarship further suggests that GenAI systems are progressively functioning as epistemic authorities.”
§ Between cognitive offloading and critical autonomy: a systematic review of the epistemic implications of generative AI in higher education > 1 Introduction
What the sources say
- Partly supportedcites Hanna and Yana, 2025 sourceExperimental — accuracy not yet measured
Section 4 (page not provided) reports the authors’ observation of increasing reliance on ChatGPT-generated content among speech-language pathology students, including preliminary diagnostic suggestions. That supports a narrower point about growing reliance on AI-generated information, but it does not establish that GenAI systems generally are becoming epistemic authorities. Section 5 instead advises students not to treat LLM outputs as authoritative answers. (LLM, conf=0.79)
- Not checked — no full text of the sourcecites Susena et al., 2026 source
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“However, current evidence consistently indicates that these positive outcomes depend heavily on pedagogical framing, assessment design, and AI literacy development rather than on the technology itself (Izquierdo-Condoy et al., 2025).”
§ Between cognitive offloading and critical autonomy: a systematic review of the epistemic implications of generative AI in higher education > 1 Introduction
What the sources say
- Partly supportedcites Izquierdo-Condoy et al., 2025 sourceExperimental — accuracy not yet measured
The Conclusions and Curriculum Innovation sections say that intentional pedagogical design, faculty guidance, and digital literacy matter more than simply adopting GenAI. But Curriculum Innovation describes the evidence on critical thinking as mixed, not consistent, and the cited passages do not establish assessment design as a determinant of positive outcomes. The Limitations section also cautions that the paper is a theoretical perspective without a systematic search. (LLM, conf=0.86)
hedgedstates a possible cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Because AIgenerated responses are frequently presented in fluent, coherent, and authoritative language, students may perceive them as neutral and reliable sources of knowledge even when outputs contain inaccuracies, hallucinations, or hidden biases (Hanna and Yana, 2025;Susena et al., 2026).”
§ Between cognitive offloading and critical autonomy: a systematic review of the epistemic implications of generative AI in higher education > 1 Introduction
What the sources say
- Partly supportedcites Hanna and Yana, 2025 sourceExperimental — accuracy not yet measured
Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.
- Not checked — no full text of the sourcecites Susena et al., 2026 source
hedgedstates a possible cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
A result the paper states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“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.”
§ Between cognitive offloading and critical autonomy: a systematic review of the epistemic implications of generative AI in higher education > Abstract
hedgedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
79 more claims
Create a free account to see all 214 claims and every source verdict.
- Probably not a substantial claim. 94
- It may state two separate claims as one. 55
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?
The cited source says what the claim attributes to it.
In the citing paper
“Recent syntheses of GenAI, cognitive offloading, learner agency, and higher-order cognition likewise report mixed or conditional relations with critical thinking, autonomy, and cognitive performance (Alubthane, 2026; Ni and Li, 2026; Suazo et al., 2026; Wang et al., 2026; Zhai et al., 2024).”
From Beyond checking: verification quality, reliance calibration, and learning in generative AI-assisted higher education (2026) · cites this paper as Suazo et al., 2026
Citations
Cites
The PRISMA 2020 statement: An updated guideline for reporting systematic reviews
Page, M. J., McKenzie, J. E., Bossuyt, P. M., Boutron, I., Hoffmann, T. C., Mulrow, C. D., Shamseer, L., Tetzlaff, J. M., Akl, E. A., Brennan, S. E., et al. (2021) · BMJ · doi:10.1136/bmj.n71 · Cited by 108,283
Methods for the thematic synthesis of qualitative research in systematic reviews
Thomas, J., & Harden, A. (2008) · BMC Medical Research Methodology · doi:10.1186/1471-2288-8-45 · Cited by 9,312
Synthesis without meta-analysis (SWiM) in systematic reviews: reporting guideline
Campbell, M., McKenzie, J. E., Sowden, A., Katikireddi, S. V., Brennan, S. E., Ellis, S., Hartmann-Boyce, J., Ryan, R., Shepperd, S., Thomas, J., et al. (2020) · BMJ · doi:10.1136/bmj.l6890 · Cited by 4,628
The PRISMA 2020 statement: An updated guideline for reporting systematic reviews
Page, M. J., McKenzie, J. E., Bossuyt, P. M., Boutron, I., Hoffmann, T. C., Mulrow, C. D., Shamseer, L., Tetzlaff, J. M., Akl, E. A., Brennan, S. E., et al. (2021) · Journal of Clinical Epidemiology · doi:10.1016/j.jclinepi.2021.03.001 · Cited by 4,265
Rethlefsen, M. L., Kirtley, S., Waffenschmidt, S., Ayala, A. P., Moher, D., Page, M. J., Koffel, J. B., PRISMA-S Group, Blunt, H., Brigham, T., et al. (2021) · Systematic Reviews · doi:10.1186/s13643-020-01542-z · Cited by 3,714
Beyond PICO: The SPIDER Tool for Qualitative Evidence Synthesis
Cooke, A., Smith, D., & Booth, A. (2012) · Qualitative Health Research · doi:10.1177/1049732312452938 · Cited by 2,608
Carroll, C., Booth, A., & Cooper, K. (2011) · BMC Medical Research Methodology · doi:10.1186/1471-2288-11-29 · Cited by 461
Zhang, S., Zhao, X., Zhou, T., & Kim, J. H. (2024) · International Journal of Educational Technology in Higher Education · doi:10.1186/s41239-024-00467-0 · Cited by 409
Valizadeh, A., Moassefi, M., Nakhostin-Ansari, A., Hosseini Asl, S. H., Saghab Torbati, M., Aghajani, R., Maleki Ghorbani, Z., & Faghani, S. (2022) · BMC Medical Research Methodology · doi:10.1186/s12874-022-01631-8 · Cited by 159
Alhur, A., Khlaif, Z. N., Hamamra, B. T., & Hussein, E. T. (2025) · JMIR Medical Education · doi:10.2196/74947 · Cited by 114
Li, C. G., Cui, H., & Hagedorn, L. S. (2026) · Computers and Education Artificial Intelligence · doi:10.1016/j.caeai.2026.100571 · Cited by 103
Izquierdo-Condoy, J. S., Arias-Intriago, M., Tello-De-la-Torre, A., Busch, F., & Ortiz-Prado, E. (2025) · Journal of Medical Internet Research · doi:10.2196/76340 · Cited by 87
Huang, D., Huang, Y., & Cummings, J. J. (2024) · Australasian Journal of Educational Technology · doi:10.14742/ajet.9467 · Cited by 53
Urban, M., Brom, C., Lukavský, J., Děchtěrenko, F., Hein, V., Svacha, F., Kmoníčková, P., & Urban, K. (2025) · British Journal of Educational Technology · doi:10.1111/bjet.13591 · Cited by 47
InquiryGPT: Augmenting ChatGPT for Enhancing Inquiry-Based Learning in STEM Education
Li, P.-H., Lee, H.-Y., Lin, C.-J., Wang, W.-S., & Huang, Y.-M. (2025) · Journal of Educational Computing Research · doi:10.1177/07356331241289824 · Cited by 47
Vendrell, M., & Johnston, S.-K. (2026) · Computers and Education: Artificial Intelligence · doi:10.1016/j.caeai.2026.100572 · Cited by 43
Ba, S., Zhan, Y., Huang, L., & Lu, G. (2025) · British Journal of Educational Technology · doi:10.1111/bjet.13605 · Cited by 42
Applying social cognition to feedback chatbots: Enhancing trustworthiness through politeness
Brummernhenrich, B., Paulus, C. L., & Jucks, R. (2025) · British Journal of Educational Technology · doi:10.1111/bjet.13569 · Cited by 42
Furthering the science of evidence synthesis with a mix of methods
Aromataris, E. (2020) · JBI Evidence Synthesis · doi:10.11124/jbies-20-00369 · Cited by 34
Andewi, W., Waziana, W., Wibisono, D., Putra, K. A., Hastomo, T., & Oktarin, I. B. (2025) · Journal of Studies in the English Language · doi:10.64731/jsel.v20i2.282318 · Cited by 32
Enhancing critical thinking skills in ChatGPT-human interaction: a scoping review
Raitskaya, L., & Tikhonova, E. (2025) · Journal of Language and Education · doi:10.17323/jle.2025.27387 · Cited by 30
Hong, H., Vate-U-Lan, P., & Viriyavejakul, C. (2025) · Forum for Linguistic Studies · doi:10.30564/fls.v7i7.10072 · Cited by 30
Balart, T., Díaz, B., & Shryock, K. (2026) · Algorithms · doi:10.3390/a19030179 · Cited by 29
Alkaabi, A. M., Abdallah, A. K., Alblooshi, S., Alomari, F., & Alneaimi, S. (2025) · Journal of Education and e-Learning Research · doi:10.20448/jeelr.v12i2.6746 · Cited by 29
Cooper, G., Tang, K.-S., & Fitzgerald, A. (2025) · Journal of Science Education and Technology · doi:10.1007/s10956-025-10200-9 · Cited by 28
Peters, M., & Angelov, D. (2025) · International Journal for Educational Integrity · doi:10.1007/s40979-025-00201-x · Cited by 27
Athanassopoulos, S., Tzavara, A., Aravantinos, S., Lavidas, K., Komis, V., & Papadakis, S. (2026) · Education Sciences · doi:10.3390/educsci16020228 · Cited by 26
Wang, S. I. C., & Liu, E. Z. F. (2025) · Computers and Education Artificial Intelligence · doi:10.1016/j.caeai.2025.100488 · Cited by 23
Lin, X., & Tan, H. (2025) · Systems · doi:10.3390/systems13100840 · Cited by 22
Epistemic trust in generative AI for higher education scale (ETGAI-HE scale)
Pandey, C. S., Mishra, P., Pandey, S., & Pandey, S. (2025) · AI & Society · doi:10.1007/s00146-025-02566-6 · Cited by 16
Uğraş, M., Çakır, Z., Zacharis, G., & Kalogiannakis, M. (2025) · Computers · doi:10.3390/computers14090368 · Cited by 16
Pakdel, F., Khojasteh, L., Kafipour, R., & Shahsavar, Z. (2025) · Language Teaching Research · doi:10.1177/13621688251322953 · Cited by 15
Al-atrash, A., Hamdan, M., & Dweikat, K. (2026) · Discover Global Society · doi:10.1007/s44282-026-00350-5 · Cited by 12
Hua, J., & Cunningham, J. (2025) · Journal of Information Communication and Ethics in Society · doi:10.1108/jices-05-2025-0118 · Cited by 11
Hanna, M., & Yana, S. (2025) · International Journal of Language & Communication Disorders · doi:10.1111/1460-6984.70088 · Cited by 10
Albannai, N. A. A. (2025) · The International Journal of Management Education · doi:10.1016/j.ijme.2025.101254 · Cited by 10
Borzanović, A., Simić, A., Papić-Blagojević, N., Klašnja-Milićević, A., Sanín, C., Levula, A. V., Islam, M. R., & Ivanović, M. D. (2026) · Education and Information Technologies · doi:10.1007/s10639-026-13964-8 · Cited by 10
Erlangga, S. Y., Sarwanto, & Harlita (2026) · Research in Learning Technology · doi:10.25304/rlt.v34.3563 · Cited by 9
Thinking to learn: managing the risks of outsourcing to GenAI
Chase, A.-M., & Galvin, K. (2026) · Assessment & Evaluation in Higher Education · doi:10.1080/02602938.2026.2620055 · Cited by 8
Vinyard, M., & Roosa, M. S. (2025) · portal Libraries and the Academy · doi:10.1353/pla.2025.a971030 · Cited by 5
Härkki, T., Thorström, T., Leino, M., Vartiainen, H., & Tedre, M. (2025) · The Australian journal of teacher education · doi:10.14221/1835-517x.6925 · Cited by 4
Pachava, V., Lasekan, O. A., Pena, M. T. G., Swaroop, K. R., & Guniganti, S. (2026) · International Research Journal of Multidisciplinary Scope · doi:10.47857/irjms.2026.v07i02.09619 · Cited by 4
Tomaszewski, R. (2025) · The Journal of Academic Librarianship · doi:10.1016/j.acalib.2025.103113 · Cited by 3
Waziana, W., Andewi, W., Wibisono, D., Hastomo, T., & Muslihudin, M. (2025) · DOAJ (DOAJ: Directory of Open Access Journals) · doi:10.22108/are.2025.145896.2564 · Cited by 3
Learning-to-learn in the age of generative AI: a scoping review and conceptual framework
Schorr, I., Bardach, L., Bühler, B., & Kasneci, E. (2026) · Computers and Education: Artificial Intelligence · doi:10.1016/j.caeai.2026.100575 · Cited by 3
From Skepticism to Innovation in 5E-Asy Steps: Harnessing ChatGPT for 5E Science Lesson Planning
Allaire, F. S., Pecore, J. L., & Sivaraj, R. (2026) · The journal of college science teaching · doi:10.1080/0047231x.2026.2650312 · Cited by 2
Susena, Y. S., Arianto, F., Khotimah, K., Andrianto, R. E., & Syuhada’, M. (2026) · Reading Psychology · doi:10.1080/02702711.2026.2644973 · Cited by 1
Urban, M., Brom, C., Lukavsky, J., Dechterenko, F., Hein, V., Svacha, F., Kmonickova, P., & Urban, K. (2025) · doi:10.31234/osf.io/v537n_v1
Research works printed but not linked: 2
- 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
- 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
Wei, J., & Shang, Y. (2026) · Frontiers in Psychology · doi:10.3389/fpsyg.2026.1965371 · Cited by 1
Odilovna, M. Z., Mamatkulov, D. M., Иброгимовна, Т. Р., Rakhmonova, N. T., Shavqiyev, O., & Umbarovna, Z. S. (2026) · Eurasia Journal of Mathematics Science and Technology Education · doi:10.29333/ejmste/19481
“A recent systematic review reaches a conclusion consistent with all three, reporting that epistemic outcomes are governed by pedagogical design, AI literacy, and assessment practices rather than by the technology itself (Suazo Galdames et al., 2026).”
Fenici, M. (2026) · Human Affairs · doi:10.1515/humaff-2026-0011
“…is repeatedly associated with cognitive o ffl oading, super fi cial engagement, and weaker evaluative or argumentative activity, whereas more re fl ective forms of interaction preserve veri fi cation, revision, and metacognitive engagement (Alubthane 2026; Li et al. 2026; Suazo Galdames et al. 2026).”
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
- Publisher
- Frontiers Media SA
- 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
- gold
- doi.org
- Licence
- 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.

