Generative AI an academic equalizer? The differential impact of AI-assisted learning on self-efficacy and intrinsic motivation among university students
Open accessThe DOI matches its Crossref record; 32 of the 32 references it cites were found in a registry or an index; 3 of the 57 sources its claims cite were checked for support.
Abstract
Generative AI tools have become more common in universities, but studies on the psychology of such use for all students remain insufficiently researched at this time point. This study investigated whether AI-assisted Learning (AIL) is associated with college Students’ intrinsic motivation and Self-efficacy, and if these relationships vary based on Academic performance. A 2 × 2 quasi-experimental design was employed with 200 STEM undergraduates, who were classified into high- and low-achievement groups and assigned to either a DeepSeek-assisted writing condition or a traditional-support writing condition. Intrinsic Motivation and Self-efficacy of Participants after completing the Scientific Paper. The data were analysed by CFA, ANOVA/ANCOVA, mediation analysis and moderated-mediation analysis. The results indicated that the post-tasks’ intrinsic motivation and self-efficacy of students under an AI-accompanied tutoring were much higher than those without it. Further analysis revealed a more pronounced impact of the Group Differences on lower achievers compared to higher ones. Also, intrinsic motivation partly explained why a Writing Support Type was associated with self-efficacy. According to the results, generative AI has equitising characteristics in higher education and provides more robust emotional assistance for weaker learners of initial knowledge.
Overview
CoChat Summary
TL;DR
In a 2 × 2 quasi-experiment with 200 STEM undergraduates, DeepSeek-assisted scientific writing was associated with higher post-task intrinsic motivation and scientific writing self-efficacy than traditional support, with larger differences among low-achieving students. Intrinsic motivation partially mediated the association between writing support type and self-efficacy.
Key findings
- DeepSeek-assisted students scored higher than traditional-support students on intrinsic motivation (M = 5.78, SD = 0.77 vs. M = 4.07, SD = 1.13) and scientific writing self-efficacy (M = 5.46, SD = 0.80 vs. M = 3.61, SD = 1.05).
- Main effects were significant for writing support type [intrinsic motivation: F(1,198) = 259.30, p < 0.001; self-efficacy: F(1,198) = 282.61, p < 0.001] and academic achievement level [F(1,198) = 28.56 and 37.92, respectively, p < 0.001]; interaction effects were also significant [F(1,198) = 35.70 and 13.46, respectively, p < 0.001]. ANCOVA controlling GPA, task familiarity, grade level, and gender retained these patterns (support type partial η² = 0.575 for intrinsic motivation and 0.595 for self-efficacy; interaction partial η² = 0.157 and 0.066).
- The support-condition advantage was larger for low-achieving students: covariate-adjusted estimated marginal means for self-efficacy were 2.703 (traditional low) vs. 4.937 (DeepSeek low), and 4.533 (traditional high) vs. 5.967 (DeepSeek high); for intrinsic motivation they were 3.136 (traditional low) vs. 5.497 (DeepSeek low), and 4.992 (traditional high) vs. 6.075 (DeepSeek high).
- Intrinsic motivation partially mediated the writing-support-type → self-efficacy association: writing support type → intrinsic motivation b = 1.717, SE = 0.137, p < 0.001; intrinsic motivation → self-efficacy b = 0.338, SE = 0.064, p < 0.001; direct effect b = 1.271, SE = 0.166, p < 0.001; indirect effect b = 0.579, BootSE = 0.116, 95% CI [0.373, 0.825].
Methods
- Design and sample: 2 × 2 quasi-experimental between-subjects design with 200 STEM undergraduates (112 boys, 88 girls; mean age 21.2 years, SD = 1.6; 48 second-year, 102 third-year, 50 fourth-year) from a scientific writing course at a Chinese university; students with prior published scientific papers were excluded. Participants were median-split by cumulative weighted GPA into high- and low-achievement groups, with 50 participants per cell.
- Procedure and conditions: 90-minute sessions involved orientation/task introduction, writing task, and post-task assessment. All participants wrote a 500–600-word IEEE-formatted abstract for a hypothetical study on a lightweight image-recognition algorithm with attention mechanism. The DeepSeek-assisted condition used DeepSeek for idea generation, content organization, literature integration, logical development, and academic expression; the traditional-support condition used academic databases and a writing guide without generative AI.
- Measures and analysis: intrinsic motivation was measured with 6 items adapted from the IMI Interest/Enjoyment subscale, and scientific writing self-efficacy with 6 items, both on a 7-point scale; pilot Cronbach’s α was 0.91 and 0.93, and main-study α was 0.958 and 0.960. Analysis used CFA in AMOS 26, MANOVA/ANOVA/ANCOVA in SPSS 27, mediation via PROCESS Model 4 with 5,000 bootstrap samples, and moderation via multiple regression.
Limitations
- Paper-stated: no pretests of intrinsic motivation or self-efficacy, so within-person change cannot be established; the quasi-experimental design was not randomized or blinded, allowing expectancy/demand effects; a novelty effect may have influenced results; self-report measures may be affected by social conformity and subjective interpretation; no objective writing-quality or academic-outcome measure was included; the sample and context limit generalizability; other mediators and boundary conditions (e.g., digital access, prompt quality, teacher guidance, AI confidence, critical AI skills) were not tested.
- Text-evident: the DeepSeek-assisted condition showed non-normal distributions for intrinsic motivation (Shapiro–Wilk p = 0.011; Kolmogorov–Smirnov p = 0.004) and self-efficacy (p = 0.027; p = 0.012), yet parametric tests were used; the very high Cronbach’s α values (0.958 and 0.960) may indicate item redundancy or high intercorrelation, as the paper itself notes.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures




Verification
Claims the paper makes
3 Valid · 0 Not valid · 31 Not checked · 97 Own results and interpretations
The sources with a verdict support this claim; the rest weren’t checked yet, have no full text we could read, or their verdicts aren’t rated yet.
In the paper
“Writing-Related Contexts: Generative AI has been associated with improved Writing Self-Efficacy via real-time assistance and process-oriented support (Pellas, 2023; Bouzar, 2024; Huang and Mizumoto, 2024).”
§ Generative AI an academic equalizer? The differential impact of AI-assisted learning on self-efficacy and intrinsic motivation among university students > Literature review > Theoretical foundations
What the sources say
- Supportedcites Pellas, 2023 source
The Implications section (point 1) states that 'the use of generative AI platforms can significantly improve undergraduate students' narrative intelligence scores and writing self-efficacy,' and point 2 attributes this to platforms 'offering personalized feedback, suggesting improvements, and facilitating the writing process' — i.e., assistance oriented to the writing process (reinforced in the Introduction by platforms that 'offer meticulous writing refinement assistance'). The Results (Table 8) confirm the effect: significant gains for the experimental group in overall writing self-efficacy (F(1,62)=4.548, p=0.039, medium-to-large η²=0.08), with 'Creative identity' the exception. The only element not spelled out in our copy is the literal phrase 'real-time'; the source describes process-facilitating, personalised feedback rather than timing, so that wording is a mild elaboration, not a contrary or unsupported component. (LLM, conf=0.82)
- Check pendingcites Bouzar, 2024 source
- Check pendingcites Huang and Mizumoto, 2024 source
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.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Systematic review and Meta-analysis indicate that generative AI may have a positive impact on the motivation and enthusiasm of students studying at higher learning institutions (Xia et al., 2025; Qian, 2025; Chen and Cheung, 2025).”
§ Generative AI an academic equalizer? The differential impact of AI-assisted learning on self-efficacy and intrinsic motivation among university students > Literature review > Generative AI and intrinsic motivation
What the sources say
- Not checked — no full text of the sourcecites Xia et al., 2025 source
- Partly supportedcites Qian, 2025 sourceExperimental — accuracy not yet measured
Qian (2025) is indeed a systematic review of GenAI in higher education (Methods, p. 3; PRISMA-guided), and it does report positive learner-facing effects close to the claim: 'GenAI tools provide significant affective support by fostering an encouraging and patient learning environment... enhances students' overall learning experiences' (Learning Support, p. 8) and 'enhancing student engagement' (Automated Feedback and Assessment, p. 8), plus 'deeper engagement... and active participation' (p. 12). However, the paper is not a meta-analysis — its synthesis is grounded coding and thematic analysis of 42 pedagogical and 25 challenge codes (Coding and Thematic Analysis, p. 3) — and it never reports 'motivation and enthusiasm' as a construct; the sole motivation finding it carries is contrary, that 'Yilmaz and Yilmaz (2023) found AI tools did not mitigate motivation gaps in programming courses' (Student AI Literacy and Dependency, p. 11). So the engagement/affective-support half of the claim is supported while the meta-analysis descriptor and the motivation-and-enthusiasm outcome are not attributable to this source. (LLM, conf=0.75)
- Not checked yetcites Chen and Cheung, 2025 source
A check is scheduled.
hedgedstates a possible cause- 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.
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Research indicates that generation AI can enhance the writing efficacy of students in their ability to revise language, organise concepts, and arrange structures better (Pellas, 2023; Bouzar, 2024; Huang and Mizumoto, 2024).”
§ Generative AI an academic equalizer? The differential impact of AI-assisted learning on self-efficacy and intrinsic motivation among university students > Literature review > Generative AI and students’ self-efficacy
What the sources say
- Partly supportedcites Pellas, 2023 sourceExperimental — accuracy not yet measured
Pellas (2023) does support the core attribution: the Results (Table 8, SAWSES ANOVA) and Discussion and Conclusions report that generative AI platforms significantly improved undergraduates' general writing self-efficacy (F = 6.54, p = 0.017) alongside narrative intelligence in story-creation tasks. However, the specific triad 'revise language, organise concepts, arrange structures' is not a measured outcome of this study — its self-efficacy factors were writing essentials, relational reflective writing and creative identity — and appears only as a forward-looking remark in Section 5 (Implications), 'These platforms can be especially useful for tasks involving grammar, structure, and organization,' which is a recommendation rather than a finding; the closest empirical detail (suggesting alternative phrasing, synonyms and sentence structures) is credited to other sources in the Introduction (refs [13]-[15]). The claim also drops the study's bounded population and setting (undergraduate Primary Education majors, web-based generative AI storytelling platforms) and states as a general research conclusion what the paper reports for that sample. (LLM, conf=0.72)
- Check pendingcites Bouzar, 2024 source
- Check pendingcites Huang and Mizumoto, 2024 source
hedgedstates a cause- 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.
- Partly supported
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The results indicated that the post-tasks’ intrinsic motivation and self-efficacy of students under an AI-accompanied tutoring were much higher than those without it.”
§ Generative AI an academic equalizer? The differential impact of AI-assisted learning on self-efficacy and intrinsic motivation among university students > 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.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Also, intrinsic motivation partly explained why a Writing Support Type was associated with self-efficacy.”
§ Generative AI an academic equalizer? The differential impact of AI-assisted learning on self-efficacy and intrinsic motivation among university students > Abstract
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.
40 more claims
Create a free account to see all 131 claims and to verify the 42 references still open.
- It may state two separate claims as one. 44
- The quote only partly states this claim. 3
- Probably not a substantial claim. 63
Statistics that don’t add up
Each reported p-value recomputed from its test statistic and degrees of freedom.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
“F(3, 576) = 4.08, p = 0.000”
§ Generative AI an academic equalizer? The differential impact of AI-assisted learning on self-efficacy and intrinsic motivation among university students > Results > Control-adjusted results
Reported p = 0 · recomputed p = .007
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.
- Doesn’t add upF(3, 576) = 4.08reported p = 0 · recomputed p = .007
“F(3, 576) = 4.08, p = .000”
- Consistentχ²(66) = 2643.316reported p < .001 · recomputed p < .001
“χ2(66) = 2643.316, p < .001”
- ConsistentF(3, 196) = 1.28reported p = .283 · recomputed p = .282
“F(3, 196) = 1.28, p = .283”
- ConsistentF(3, 196) = 0.85reported p = .467 · recomputed p = .468
“F(3, 196) = 0.85, p = .467”
- ConsistentF(1, 198) = 282.61reported p < .001 · recomputed p < .001
“F(1, 198) = 282.61, p < .001”
12 more recomputed tests
Create a free account to see every test the Stats check recomputed.
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.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Generative AI an academic equalizer? The differential impact of AI-assisted learning on self-efficacy and intrinsic motivation among university students > Method > Procedure
“Ask them to write a 500-600-word, IEEE-formatted abstract for a hypothetical study on developing a lightweight image-recognition algorithm with attention mechanism capabilities in their own research project.”
§ Generative AI an academic equalizer? The differential impact of AI-assisted learning on self-efficacy and intrinsic motivation among university students > Method > Strategies
“Sample item: “I found the process of writing this Introduction to be enjoyable.””
Citations
Cites
Ryan, R. M., & Deci, E. L. (2000) · American Psychologist · doi:10.1037//0003-066x.55.1.68 · Cited by 54,599
The "what" and "why" of goal pursuits: Human needs and the self-determination of behavior
Deci, E. L., & Ryan, R. M. (2000) · Psychological Inquiry · doi:10.1207/s15327965pli1104_01 · Cited by 34,210 · cited 2×
“…Theory, extrinsic motivation tends to arise in the absence of students’ fundamental psychological needs of autonomy, competence and sense of belonging (Deci and 10.3389/fpsyg.2026.1779741 Frontiers in Psychology 03 frontiersin.org Ryan, 2000; Ryan and Deci, 2000; Deci and Ryan, 2013).”
InfluentialCognitive Load During Problem Solving: Effects on Learning
Sweller, J. (1988) · Cognitive Science · doi:10.1016/0364-0213(88)90023-7 · Cited by 11,048
Cognitive Load Theory and Instructional Design: Recent Developments
Paas, F., Renkl, A., & Sweller, J. (2003) · Educational Psychologist · doi:10.1207/s15326985ep3801_1 · Cited by 3,041
Impact of Self-Regulatory Influences on Writing Course Attainment
Zimmerman, B. J., & Bandura, A. (1994) · American Educational Research Journal · doi:10.2307/1163397 · Cited by 1,237
Yilmaz, R., & Karaoglan Yilmaz, F. G. (2023) · Computers and Education: Artificial Intelligence · doi:10.1016/j.caeai.2023.100147 · Cited by 745
Jin, Y., Yan, L., Echeverria, V., Gašević, D., & Martinez-Maldonado, R. (2025) · Computers and Education Artificial Intelligence · doi:10.1016/j.caeai.2024.100348 · Cited by 452
Jin, Y., Yan, L., Echeverria, V., Gašević, D., & Martinez-Maldonado, R. (2024) · arXiv · doi:10.48550/arxiv.2405.11800 · Cited by 447
Pedagogical Applications of Generative AI in Higher Education: A Systematic Review of the Field
Qian, Y. (2025) · TechTrends · doi:10.1007/s11528-025-01100-1 · Cited by 206 · cited 2×
“Mohammed reported enhanced tasks-driven abilities due to the training programme (Mohamed et al., 2024 Areas of Study at the tertiary level (Qian, 2025; Chen and Cheung, 2025).”
Zhang, L., & Xu, J. (2025) · The Internet and Higher Education · doi:10.1016/j.iheduc.2024.100978 · Cited by 168
Research works printed but not linked: 3
- Bandura, 1997, Self-Efficacy: The Exercise of Control
- Deci, 2013, Intrinsic Motivation and Self-Determination in Human Behavior
- Tanveer, 2024, Examining the impact of AI based Chatbots on academic self-efficacy and self-regulation among university students, J. Dev. Soc. Sci., 5, 468, 10.47205/jdss.2024(5-ii-s)45 · doi:10.47205/jdss.2024(5-ii-s)45
Cited by
Su, J., Wang, Z., Qiang, Z., & Wang, Z. (2026) · Journal of Intelligence · doi:10.3390/jintelligence14080182 · Cited by 9
Zhao, Y., Chen, J., Dai, W., & Gu, Y. (2026) · Frontiers in Psychology · doi:10.3389/fpsyg.2026.1878514 · Cited by 3
Rasoli, U., & Smiju, I. S. (2026) · Journal of Interdisciplinary Studies in Education · doi:10.32674/ppgcph83
Rahim, S., Ziauddin, Khalil, S. Z., Naoreen, B., & Shahzad, S. (2026) · Aposta · doi:10.23900/ra.v24i114.1556
Mansız, A. S. (2026) · International Journal of Technology in Education · doi:10.46328/ijte.9187
Metadata
- Authors
- Shao, H., & Wang, R.
- Affiliations
- Guangxi Science & Technology Normal University (CN)
- Corresponding author
- Wang, Rifeng
- ORCID
- Wang, Rifeng
- Venue
- Frontiers in Psychology
- Publisher
- Frontiers Media SA
- Published
- 2026-03-30
- Volume
- 17
- Pages
- 1779741-1779741
- ISSN
- 1664-1078 (ISSN-L)
- PMC
- PMC13071819
- OpenAlex
- W7143420043
- Cited by
- 14
- References
- 31
- Citations by year
- 2026: 14
- FWCI
- 54.59
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- Frontiers in Psychology (publishedVersion, cc-by)
- PubMed Central (publishedVersion, cc-by)
- PubMed (publishedVersion)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Topics
- Social Sciences › Education › Artificial Intelligence in Education
- Social Sciences › Psychology › Developmental and Educational Psychology › Education, Achievement, and Giftedness
- Physical Sciences › Computer Science › Artificial Intelligence › Intelligent Tutoring Systems and Adaptive Learning
- Keywords
- generative AI, AI-assisted learning, self-efficacy, intrinsic motivation, academic performance, academic equity, writing support, DeepSeek, STEM undergraduates, academic writing, lower-achieving students
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 2026-11-18, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-11-18 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (qwen3.8-omni-flash).
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: Generative AI an academic equalizer? The differential impact of AI-assisted learning on self-efficacy and intrinsic motivation among university students. https://app.cochat.ai/papers/10.3389/fpsyg.2026.1779741
- The paper
- Shao, H., & Wang, R. (2026). Generative AI an academic equalizer? The differential impact of AI-assisted learning on self-efficacy and intrinsic motivation among university students. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1779741
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.