The DOI matches its Crossref record; 12 of the 12 references it cites were found in a registry or an index; 15 of the 27 sources its claims cite were checked for support.
Distinguishing performance gains from learning when using generative AI
Open accessAbstract
Generative artificial intelligence (AI) is increasingly being integrated into education, where it can boost learners’ performance. However, these uses do not promote the deep cognitive and metacognitive processing that are required for high-quality learning.
Overview
CoChat Summary
TL;DR
This commentary argues that generative AI tools can boost learners' observable task performance while undermining the cognitive and metacognitive processes needed for durable learning. The authors call for research designs that separate performance from learning, using retention and transfer assessments rather than immediate task-success metrics.
Key findings
- The paper distinguishes performance (observable behaviour during a task, often externally supported) from learning (enduring, transferable change from experience), noting that high performance does not imply learning — e.g., cramming before a test.
- It argues this distinction is conflated in current research: a meta-analysis of 69 studies reported that ChatGPT and similar tools improve student "academic performance" with an effect of Hedges' g = 0.7, which the authors say probably reflects immediate task success rather than learning.
- Another cited study found undergraduate peer-feedback performance improved with generative AI assistance, but the gains "diminished once the generative AI assistance was removed."
- Generative AI may reduce cognitive load (students using ChatGPT for socio-scientific research reported significantly lower cognitive load than those using traditional search engines) but this does not guarantee deeper engagement; heavy ChatGPT reliance was linked to weaker argumentation and reasoning skills.
- A randomized controlled study cited found frequent generative AI use in argumentative essay writing can foster "metacognitive laziness" in undergraduates, and frequent chatbot interaction was associated with lower learning autonomy; frequent use also raised perceived confidence and efficiency while increasing technological dependence.
Methods
- This is a conceptual commentary/opinion piece in Nature Reviews Psychology (accepted manuscript, 4, 435–436); it presents no primary data collection or analysis of its own.
- The argument is built by synthesizing and interpreting findings from cited prior studies (including a 69-study meta-analysis, a randomized controlled trial, and survey/observational studies) alongside cognitive psychology frameworks: cognitive load theory, self-determination theory, and the learning-versus-performance literature.
- The authors propose a research agenda: use process-oriented assessments such as retention and transfer tests and delayed recall tasks, investigate encoding/consolidation/retrieval mechanisms, and study longitudinal effects of educational generative AI.
Limitations
- The paper states no formal limitations section; as a short commentary it does not report its own sample, data, or analyses.
- Evident from the text: all empirical claims are drawn from other studies rather than generated here, and the cited evidence is described at a high level without reporting the primary studies' designs, sample sizes, or effect sizes beyond the single g = 0.7 figure.
- Evident from the text: the proposed research agenda (retention/transfer tests, longitudinal designs) is a set of recommendations, not tested in this work.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
12 Valid · 1 Not valid · 10 Not checked · 9 Own results and interpretations
Some cited sources support this claim and some don’t — read each source’s verdict below.
In the paper
“Generative artificial intelligence (AI) is increasingly being integrated into education, where it can boost learners' performance.”
- Also stated in: § Main · p. 2 · plain
“although generative AI can boost learners' performance on tasks, it risks undermining the cognitive and metacognitive processes that are essential for durable learning[3, 4].”
- Also stated in: § Outlook · p. 4 · plain
“Although generative AI tools such as ChatGPT can enhance learners' performance and reduce cognitive load, they might disrupt the deeper processes of encoding, retention and independent problem-solving that are necessary for durable learning.”
What the sources say
- Disputedcites [3] Stadler, M., Bannert, M. & Sailer, M. Cognitive ease at a cost: LLMs reduce mental effort but compromise depth in student scientific inquiry. Comput. Hum. Behav. 160, 108386 (2024). sourceExperimental — accuracy not yet measured
The study reports the opposite result for the learning task it tested: students using ChatGPT produced lower-quality reasoning and argumentation than students using Google (Abstract, p. 1). The Discussion (p. 5) likewise says the web-search group performed better on the quality of its justifications, although the LLM group experienced lower cognitive load. (LLM, conf=0.84)
Checked by a reasoning model — may be wrong.
- Supportedcites [4] Fan, Y. et al. Beware of metacognitive laziness: effects of generative artificial intelligence on learning motivation, processes, and performance. Br. J. Educ. Technol. 56, 489–530 (2024). source
The Practitioner Notes (p. 3) state that ChatGPT can significantly improve short-term task performance. This supports the claim that generative AI can boost learners’ performance on tasks, though the paper does not claim that the benefit extends to knowledge gain or transfer. (LLM, conf=0.94)
plainstates 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.
- Also stated in: § Main · p. 2 · plain
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Offloading metacognitive activities to generative AI can also stunt intrinsic motivation, as the learner's role in task completion diminishes[4].”
What the sources say
- Partly supportedcites [4] Fan, Y. et al. Beware of metacognitive laziness: effects of generative artificial intelligence on learning motivation, processes, and performance. Br. J. Educ. Technol. 56, 489–530 (2024). sourceExperimental — accuracy not yet measured
The discussion of self-regulated learning (p. 18) says GenAI may take over regulation from learners and reduce their engagement in metacognitive tasks, supporting the offloading part of the claim. But the intrinsic-motivation discussion (p. 17) reports no significant differences between groups and treats long-term effects on motivation as an open question. The paper therefore does not establish that offloading stunts intrinsic motivation. (LLM, conf=0.88)
plainstates 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.
- Partly supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“Generative AI tools can in some cases reduce cognitive load, particularly for tasks that require complex information processing.”
- Also stated in: § Outlook · p. 4 · plain
“Although generative AI tools such as ChatGPT can enhance learners' performance and reduce cognitive load, they might disrupt the deeper processes of encoding, retention and independent problem-solving that are necessary for durable learning.”
What the sources say
- Supportedcites [3] Stadler, M., Bannert, M. & Sailer, M. Cognitive ease at a cost: LLMs reduce mental effort but compromise depth in student scientific inquiry. Comput. Hum. Behav. 160, 108386 (2024). source
The Discussion (p. 5) describes the study’s task as a complex socio-scientific issue and reports lower cognitive load for students using an LLM than for those using traditional web searches. The claim’s qualified wording reflects that finding. (LLM, conf=0.91)
plainstates 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.
- Also stated in: § Outlook · p. 4 · plain
Every source the paper cites for this claim was checked and supports it.
In the paper
“For instance, students who used ChatGPT to research socio-scientific issues reported significantly lower cognitive load compared with those who used more traditional research tools such as internet search engines[3].”
What the sources say
- Supportedcites [3] Stadler, M., Bannert, M. & Sailer, M. Cognitive ease at a cost: LLMs reduce mental effort but compromise depth in student scientific inquiry. Comput. Hum. Behav. 160, 108386 (2024). source
The study setup (Section 5.1.1, p. 3) compares students assigned to ChatGPT 3.5 with students assigned to Google search. The task involved researching nanoparticles in sunscreen, a socio-scientific issue (Section 4, p. 3), and the Results (Section 6.1.1, p. 5) report substantially lower cognitive load in the ChatGPT group across all three measured facets. (LLM, conf=0.94)
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“This risk is supported by evidence showing that undergraduate students who relied heavily on ChatGPT for information gathering exhibited weaker argumentation and reasoning skills compared with those who used traditional, cognitively demanding research methods[3].”
What the sources say
- Partly supportedcites [3] Stadler, M., Bannert, M. & Sailer, M. Cognitive ease at a cost: LLMs reduce mental effort but compromise depth in student scientific inquiry. Comput. Hum. Behav. 160, 108386 (2024). sourceExperimental — accuracy not yet measured
The Results (section 6.1.1, p. 5) report that students assigned to web search gave significantly more relevant arguments than those assigned to ChatGPT, and that the ChatGPT group experienced lower cognitive load. The paper describes participants as university students (section 5.1, p. 3), not specifically undergraduates. It measured the quality of arguments in one task, not participants’ general reasoning skills or how heavily they relied on ChatGPT. (LLM, conf=0.86)
plainstates an association- 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
21 more claims · 7 with their cited sources checked
Create a free account to see all 32 claims and every source verdict.
- Probably not a substantial claim. 4
- It may state two separate claims as one. 3
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 source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Experimental evidence further suggests that support emphasizing direct answer provision can reduce generative cognitive engagement and cognitive activation, sometimes described as 'metacognitive laziness' (Fan et al., 2025; see also Yan et al., 2025).”
From Enhancing School Students' Self-Regulated Learning through Generative AI Support: A Randomized Controlled Trial (2026) · cites this paper as Yan et al., 2025
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“This model helps reconcile the “performance–learning paradox” noted in recent empirical and review work (Yan et al., 2025; Denny et al., 2024).”
From Tool, tutor, or crutch?: A grounded theory of cognitive scaffolding and offloading in AI-assisted programming education (2026) · cites this paper as Yan et al., 2025
The cited source says what the claim attributes to it.
In the citing paper
“Tool-supported performance must also be distinguished from durable capability development: performance gains produced by GenAI do not necessarily constitute learning or skill acquisition (Fauth and González-Martínez, 2021; Melumad and Yun, 2025; Yan et al., 2025).”
From Responsible use of generative AI in Chinese vocational education: eye-tracking evidence on source verification and immediate near-transfer across three information-seeking conditions (2026) · cites this paper as Yan et al., 2025
The check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“This form of selection-anchored dialogue addresses a limitation identified in research on unstructured AI-supported dialogue: students who already know what to ask and how to interpret responses benefit disproportionately from open-ended conversational access, while students with less developed feedback literacy may use the dialogue superficially or not at all [24, 41].”
From Making AI-Generated Feedback Matter: A Large-Scale Study of Feedback Workflows and Student Enactment (2026) · cites this paper as [41]
Experimental — accuracy not yet measuredIn the citing paper
“Most studies focus on explicit AI assistance and outcome-oriented metrics, often neglecting the temporal dynamics and processual nature of the interaction (Yang et al. 2025;Yan, Greiff, et al. 2025;Molenaar 2022b).”
From Emergent Learner Agency in Implicit Human– AI Collaboration: How Supportive and Contrarian AI Personas Reshape Interaction (2026) · cites this paper as Yan, Greiff, et al. 2025
Citations
Cites
Ryan, R. M., & Deci, E. L. (2020) · Contemporary Educational Psychology · doi:10.1016/j.cedpsych.2020.101860 · Cited by 6,037 · cited 2×
“Specifically, balancing cognitive load, metacognition and self-efficacy are critical to achieving deeper, lasting learning[5, 7, 8].”
Sweller, J., Ayres, P., & Kalyuga, S. (2011) · doi:10.1007/978-1-4419-8126-4 · Cited by 5,531
Sweller, J. (2011) · Psychology of Learning and Motivation · doi:10.1016/b978-0-12-387691-1.00002-8 · Cited by 4,856
Sweller, J. (2012) · Encyclopedia of the Sciences of Learning · doi:10.1007/978-1-4419-1428-6_446 · Cited by 4,405
Zhai, C., Wibowo, S., & Li, L. D. (2024) · Smart Learning Environments · doi:10.1186/s40561-024-00316-7 · Cited by 1,567 · cited 2×
“Along these lines, undergraduate learners who interacted frequently with generative-AI-powered chatbots for acquiring academic domain-specific knowledge reported lower learning autonomy relative to students who had infrequent chatbot interactions[9].”
Fan, Y., Tang, L., Le, H., Shen, K., Tan, S., Zhao, Y., Shen, Y., Li, X., & Gašević, D. (2025) · British Journal of Educational Technology · doi:10.1111/bjet.13544 · Cited by 805 · cited 3×
“Offloading metacognitive activities to generative AI can also stunt intrinsic motivation, as the learner’s role in task completion diminishes[4].”
Self-citationLearning Versus Performance: An Integrative Review
Soderstrom, N. C., & Bjork, R. A. (2015) · Perspectives on Psychological Science · doi:10.1177/1745691615569000 · Cited by 651
Impact of AI assistance on student agency
Darvishi, A., Khosravi, H., Sadiq, S., Gašević, D., & Siemens, G. (2024) · Computers & Education · doi:10.1016/j.compedu.2023.104967 · Cited by 548 · cited 1×
“Another study identified significant boosts to undergraduate student performance in providing quality peer feedback when using generative AI-based assistance[6].”
Self-citationDoes ChatGPT enhance student learning? A systematic review and meta-analysis of experimental studies
Deng, R., Jiang, M., Yu, X., Lu, Y., & Liu, S. (2025) · Computers & Education · doi:10.1016/j.compedu.2024.105224 · Cited by 531 · cited 2×
“Celebrated for providing instant assistance and personalized feedback, these tools promise to enhance performance across tasks such as academic writing, computer programming and information research[1, 2].”
Promises and challenges of generative artificial intelligence for human learning
Yan, L., Greiff, S., Teuber, Z., & Gašević, D. (2024) · Nature Human Behaviour · doi:10.1038/s41562-024-02004-5 · Cited by 499 · cited 1×
“Celebrated for providing instant assistance and personalized feedback, these tools promise to enhance performance across tasks such as academic writing, computer programming and information research[1, 2].”
Self-citation
Cited by
Yan, L., Pammer‐Schindler, V., Mills, C., Nguyen, A., & Gašević, D. (2025) · British Journal of Educational Technology · doi:10.1111/bjet.70000 · Cited by 84
Self-citationYuriev, E., Orgill, M., & Holme, T. A. (2025) · Journal of Chemical Education · doi:10.1021/acs.jchemed.5c01105 · Cited by 33
Wong, S. S. H., & Qiu, S. X. (2026) · Educational Psychology Review · doi:10.1007/s10648-026-10118-7 · Cited by 21
Seung, Y., & Basham, J. D. (2026) · Learning Disability Quarterly · doi:10.1177/07319487261439132 · Cited by 12
Presenting your AI in education research with rigor: The TEP-AIED model
Hwang, G.-J., Xie, H., Wah, B. W., & Gašević, D. (2026) · Computers and Education: Artificial Intelligence · doi:10.1016/j.caeai.2026.100598 · Cited by 9
Self-citation
Metadata
- Authors
- Yan, L., Greiff, S., Lodge, J. M., & Gašević, D.
- Affiliations
- Monash University (AU)
- Technical University of Munich (DE)
- The University of Queensland (AU)
- Corresponding author
- Yan, Lixiang
- Venue
- Nature Reviews Psychology
- Published
- 2025-06-18
- Volume
- 4
- Issue
- 7
- Pages
- 435-436
- ISSN
- 2731-0574 (ISSN-L)
- OpenAlex
- W4411430598
- Semantic Scholar
- 9c78fdf565f6f596575193aba03b4a6110beeb1e
- Cited by
- 63
- References
- 11
- Citations by year
- 2026: 56 · 2025: 7
- FWCI
- 64.93
- Type
- journal article
- Language
- en
- Open access
- green
- arxiv.org
- Licence
- www.springernature.com/gp/researchers/text-and-data-mining (tdm)
- www.springernature.com/gp/researchers/text-and-data-mining (vor)
- Locations
- Nature Reviews Psychology (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- HAL (submittedVersion)
- arXiv (Cornell University)
- Topics
- Social Sciences › Education › Artificial Intelligence in Education
- Physical Sciences › Computer Science › Artificial Intelligence › Intelligent Tutoring Systems and Adaptive Learning
- Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
- Keywords
- generative AI, artificial intelligence in education, metacognition, educational technology, learner performance
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 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (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: Distinguishing performance gains from learning when using generative AI. https://app.cochat.ai/papers/10.1038/s44159-025-00467-5
- The paper
- Yan, L., Greiff, S., Lodge, J. M., & Gašević, D. (2025). Distinguishing performance gains from learning when using generative AI. Nature Reviews Psychology. https://doi.org/10.1038/s44159-025-00467-5
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.