Skip to content
Review Open access

AI tool use, self-efficacy, foreign language enjoyment, and willingness to communicate: a moderated serial mediation model among Chinese EFL learners

Jul 2026 · Frontiers in Psychology · Vol 17 · 0 citations · 54 references
Medicine

TL;DR

A moderated serial mediation model in which AI tool use predicts willingness to communicate sequentially through AI self-efficacy and foreign language enjoyment (FLE) is proposed and tested, with foreign language anxiety (FLA) moderating the FLE-to-WTC link.

Abstract

Introduction The rapid proliferation of artificial intelligence (AI) tools in higher education has transformed the landscape of foreign language instruction, yet the mechanisms through which AI tool use shapes learners' affective and communicative engagement remain poorly understood. Drawing on social cognitive theory (SCT) and control-value theory (CVT), this study proposes and tests a moderated serial mediation model in which AI tool use predicts willingness to communicate (WTC) sequentially through AI self-efficacy and foreign language enjoyment (FLE), with foreign language anxiety (FLA) moderating the FLE-to-WTC link. Methods Survey data were collected from 420 EFL undergraduates at two Chinese universities (Zhejiang and Shaanxi Provinces). Covariance-based structural equation modeling (CBSEM) with maximum likelihood robust (MLR) estimation in Mplus 8.8 was used. Results The model revealed excellent fit. AI tool use significantly predicted AI self-efficacy (β = 0.42, p < 0.001), which predicted FLE (β = 0.38, p < 0.001) and WTC directly (β = 0.21, p = 0.001); FLE predicted WTC (β = 0.33, p < 0.001). Bootstrapped mediation analyses (10,000 replications) confirmed a simple indirect effect via self-efficacy [β = 0.091, 95% CI (0.042, 0.152)] and a serial indirect effect via self-efficacy and enjoyment [β = 0.053, 95% CI (0.021, 0.098)]. FLA moderated the FLE-to-WTC pathway (β = −0.19, p = 0.002), and the index of moderated mediation was significant [IMM = −0.017, 95% CI (−0.038, −0.004)]. Discussion These findings advance theoretical understanding of how AI tools cultivate positive affective trajectories and carry implications for emotionally responsive EFL pedagogy.

Read PDF

Similar papers

Review Open access Jul 2026

When ease of use is not enough: self-efficacy and continued use of AI educational tools in resource-constrained universities

Examination of college students’ continuance intention to use AI educational tools in local undergraduate universities across five northwestern Chinese provinces and a capability-centric extension of the unified theory of acceptance and use of technology (UTAUT) are developed.

Yao Wan, Kuan-Min Lu · 0 citations
Open access Aug 2026

Boundary Conditions of "More Use, Better Outcome": The Role of Self-Regulation in AI-Assisted Oral Proficiency

The rapid development of generative AI (GenAI) technology has led to the widespread adoption of AI oral practice tools among university EFL (English as a Foreign Language) learners, yet the boundary conditions under which such tools improve oral proficiency remain underexplored. This study aims to investigate whether the assumption that more frequent use leads to better outcomes is universally valid or is conditioned by learners’ AI learning engagement willingness. Based on questionnaire responses from 248 university students across multiple institutions in China, this study examined the relationships among AI oral tool usage frequency, AI learning engagement willingness (comprising task value perception, willingness to invest resources, and AI-augmented self-efficacy), and self-reported English oral proficiency. Statistical analyses, including one-way ANOVA and hierarchical multiple regression, revealed that both AI learning engagement willingness and usage frequency were independently and positively associated with oral proficiency, displaying an additive rather than interactive relationship. These findings reframe the more use, better outcome logic by revealing a dual-path mechanism: usage frequency ensures the floor of language exposure, while AI learning engagement willingness determines the ceiling of deep processing. The study provides theoretical and practical implications for AI tool design and pedagogical implementation in college English education.

Shijie Hu, Yu-Yang Zhao, Yu-Han Zeng et al. · 0 citations
Conference Open access 2026

Understanding AI-Assisted EFL Learning: The Roles of Expectancy Value, AI Literacy, and Self-Regulated Learning [Abstract]

Aim/Purpose This study investigated how learner-related factors: expectancy value, AI literacy, and self-regulated learning influence university students’ engagement in AI-assisted EFL learning environments. Background While AI-assisted language learning offers personalized and adaptive learning opportunities, its effectiveness depends on learners’ motivation, technological understanding, and ability to regulate their own learning. Methodology A survey research design was employed using a structured questionnaire developed from relevant literature and pilot tested to ensure validity and reliability. The sample consisted of 322 university students from eight universities across northern, central, southern, and eastern Taiwan. Data were analyzed using descriptive statistics, independent-samples t-tests, and one-way ANOVA. Contribution This study provides empirical evidence on how expectancy value, AI literacy, and self-regulated learning jointly shape student engagement in AI-assisted EFL contexts, extending expectancy-value theory into technology-enhanced language learning. Findings Results indicate that students reported relatively high levels of AI literacy, expectancy value, and self-regulated learning. Among sub-dimensions, AI ethics scored the highest, while resource management scored the lowest. No significant gender differences were found. However, significant differences emerged across English proficiency levels for all three variables, suggesting that perceived proficiency plays a key role in shaping motivation, AI literacy, and learning regulation. Recommendations for Practitioners Educators should design AI-assisted EFL activities that enhance learners’ motivation, develop AI literacy (especially ethical awareness), and support self-regulated learning strategies, particularly in resource management. Recommendations for Researchers Future studies should explore causal relationships among these variables, incorporate longitudinal designs, and examine how instructional interventions can strengthen AI literacy and self-regulated learning in diverse contexts. Impact on Society By improving understanding of learner factors in AI-assisted education, this study supports the development of more effective and equitable technology-enhanced language learning environments. Future Research Further research should investigate how different types of AI tools influence learning outcomes, as well as how individual differences such as proficiency, motivation, and digital competence interact over time in AI-supported learning contexts.

Kate Tzu-Ching Chen, Ming-Tzer Lin · 0 citations
Open access Jul 2026

Technical Proficiency as the Strongest Predictor of AI Self-Efficacy: A Decision Tree Analysis of ChatGPT Literacy Among Romanian Educators

The growing integration of generative artificial intelligence (AI) in education requires educators to develop not only operational competencies but also confidence in using AI tools effectively. The present study examined the predictive relationship between the multidimensional construct of ChatGPT literacy and AI self-efficacy among educators working in Romanian educational settings. A sample of 393 educators from Western Romania completed the ChatGPT Literacy Scale (and the AI Self-Efficacy subscale of the Meta AI Literacy Scale. Reliability analyses indicated good to excellent internal consistency across all literacy dimensions (α = .80–.95) and AI self-efficacy (α = .89). To model nonlinear and hierarchical relationships among predictors, a Decision Tree Regression approach was implemented in JASP. The five ChatGPT literacy dimensions, technical proficiency, critical evaluation, communication proficiency, ethical competence, and creative application, were entered as predictors of AI self-efficacy. The model explained 53.2% of the variance in AI self-efficacy (R² = .532), demonstrating moderate predictive performance (MSE = 0.517; RMSE = 0.719). Feature importance analysis revealed that technical proficiency was the strongest predictor (40.07%), followed by ethical competence (18.12%), critical evaluation (14.99%), communication proficiency (14.01%), and creative application (12.80%). The first and most informative split occurred on technical proficiency, highlighting its importance in shaping educators’ perceived AI capability. These findings indicate that technical proficiency emerged as the strongest predictor of AI self-efficacy among educators. The results have implications for AI-focused professional development programmes, emphasising the importance of structured technical training alongside ethical and critical competencies.

Ioana-Eva Cădariu, Loredana-Ileana Vîșcu, Cristian Delcea et al. · 0 citations
Open access Aug 2026

Latent Profiles of AI Literacy, Self-Directed Learning, and Creative Self-Efficacy among Art Students and Their Associations with Lifelong Learning Competence

Generative artificial intelligence (AI) is reshaping the resources art students use for learning and creative production, but technological knowledge, learning regulation, and creative confidence may not develop in parallel. This cross-sectional study used latent profile analysis to examine how self-reported AI literacy (AIL), self-directed learning (SDL), and a brief four-item contextualized measure of creative self-efficacy (CSE) co-occurred among 502 art students from three universities in Shaanxi Province, China. Four profiles were retained as a descriptive representation of the sample: moderate AIL-lower SDL/CSE (15.7%), lower AIL-moderate SDL/CSE (18.7%), higher AIL-moderate SDL/CSE (32.5%), and higher AIL-higher SDL/CSE (33.1%). CSE item loadings ranged from .673 to .724, with alpha = .785, omega = .786, and average variance extracted = .479, indicating adequate internal consistency but limited content coverage and marginal convergent validity. Replacing the four CSE items with their composite mean produced high classification agreement with the primary model (adjusted Rand index [ARI] = .869); removing CSE reduced agreement (ARI = .729) but retained substantial heterogeneity. Neither sensitivity specification uniquely supported four rather than three profiles. Prior AI-course experience distinguished the first profile from the higher AIL-higher SDL/CSE profile, and self-reported lifelong learning competence increased across the four profiles. The findings support flexible course designs, but the profiles remain sample- and specification-bound and should not be used for individual screening.

Huanxiang Gao, Wei Wang, Ningning Wang · 0 citations
Open access Aug 2026

The effects of perceived technical explanation and analogical explanation on learning behavioral intention among K-12 students with different subject preferences in AI-assisted learning.

Preliminary evidence that subject preference conditions the effectiveness of technical explanation on behavioral intention is offered, suggesting that AIE style differentiation warrants consideration in GAI system design for K-12 learners.

Zhou Jin, Zhongtian Lv, Yingxin Li et al. · 0 citations