Impacts of artificial intelligence on willingness to communicate and its key antecedents in EFL contexts: a systematic review (2015–2026)

Synthesis: Zhang, Zou, Liu, Chen, and Yan (2026) synthesize 73 empirical studies published between January 2015 and February 2026 to ask how artificial intelligence influences EFL learners' willingness to communicate (WTC) and its key antecedents. Searching six databases, they found a young and rapidly expanding field: no studies before 2021, 52 of the 73 published in 2025 alone, and a strong concentration in Asian EFL settings and on generative AI chatbots. The review argues that AI supports willingness to communicate directly, since every study that measured WTC as an outcome reported positive effects, and indirectly through four proposed pathways: Affective Foundation (anxiety reduction), Motivational Propensity, Communicative Capacity, and Contextual Affordance. Anchored in MacIntyre et al.'s pyramid model, the framework reframes AI as a multi-layered affordance for communication. The authors also record the counter-evidence, including technical instability, negative emotional impacts, over-reliance on AI, and context-dependence, and close with five research gaps and a call for longitudinal, triangulated, and theory-driven work.

Key Findings

1. 73 empirical studies published between January 2015 and February 2026 were synthesized; 52 (71.2%) appeared in 2025, and none before 2021.
2. Every study that measured willingness to communicate as a primary outcome reported positive effects of AI; 26 studies were mapped to direct promotion of WTC.
3. The review proposes four indirect pathways: Affective Foundation (anxiety reduction), Motivational Propensity, Communicative Capacity, and Contextual Affordance, anchored in MacIntyre et al.'s pyramid model.
4. Counter-evidence is coded separately: technical instability in 8 studies, negative emotional impacts in 4, over-reliance on AI in 3, and contextual contingency in 1.
5. Quality appraisal with the MMAT rated 28 studies (38.4%) high, 39 (53.4%) medium, and 6 (8.2%) low quality across 38 mixed-methods, 31 quantitative, and 4 qualitative designs.
6. The review identifies five interrelated gaps (methodological, contextual, population, technological, theoretical) and calls for longitudinal designs, triangulated WTC measures, and comparative tool research.

Willingness to communicate and its antecedents

Willingness to communicate is a learner's readiness to engage in communication in the target language at a specific moment with a particular person or group, and it often decides whether a learner speaks or stays silent. In EFL settings, higher WTC travels with stronger communication skills and greater willingness to take risks in authentic interaction. The review focuses on three antecedents identified as central: foreign language anxiety, self-confidence and self-efficacy, and Motivation, with Student Engagement as a closely related factor. AI is expected to act on them by offering private, low-pressure practice free of judgment, personalized feedback with visible progress, and gamified features that sustain participation. The framing follows MacIntyre et al.'s pyramid model of L2 WTC, extended to AI-mediated learning.

How the review was conducted

This PRISMA-aligned systematic review searched six databases spanning applied linguistics, education, psychology, and computer science, using three keyword sets: WTC and its key antecedents, AI technology, and EFL contexts. From 556 records, duplicate removal left 368; title and abstract screening excluded 289; 75 full texts were assessed and 73 studies were included, with high agreement at every stage. Mixed-methods designs dominate the corpus (38 studies), followed by quantitative (31) and qualitative (4). Quality was appraised with the Mixed Methods Appraisal Tool, and no study was excluded on quality grounds.

Four pathways from AI to willingness to communicate

The review's central contribution is an integrated framework. Studies that measured WTC as a primary outcome reported positive effects of AI, read as direct promotion of willingness to communicate (26 studies). Around that direct effect the authors synthesized four indirect pathways. The Affective Foundation pathway, mapped in 46 studies, works through anxiety reduction: AI offers private, low-pressure practice that removes fear of mistakes and negative evaluation. The Motivational Propensity pathway covers gamified and interactive features that sustain motivation for oral practice. The Communicative Capacity pathway, mapped in 28 studies, runs through speaking proficiency, where corrective feedback and repeated practice build oral skills. The Contextual Affordance pathway covers tailored feedback and Scaffolding (20 studies) plus Simulation of authentic interaction that increases target language exposure (14 studies). The authors caution that these are theoretical syntheses, since only a small subset has been tested with formal mediation, and that their strength is moderated by AI tool design, learners' AI literacy, and teacher support. Technical instability, cognitive overload, and over-reliance on AI constrain the picture.

What this means for practice

• Instructors. Use AI practice as low-stakes rehearsal that lowers fear of mistakes, then bridge it to classroom speaking tasks.
• Instructors. Favor tools that give personalized feedback and show progress, and test speech recognition accuracy before relying on it for oral work.
• Developers and designers. Report implementation details (interface, prompt structure, voice functions, feedback configuration) and compare tools head to head, because controlled comparative evidence is scarce.
• Researchers. Triangulate self-report WTC scales with observed behavior and interaction logs, and test the four pathways with formal mediation rather than assuming a linear path.

Limitations

• The synthesis is thematic and descriptive; the authors state it cannot capture causal relationships, and most pathways lack formal mediation tests.
• The search covered only English-language peer-reviewed empirical studies, and the EFL-only scope limits transfer to ESL settings.
• The antecedent keyword set was limited to anxiety, self-confidence, and motivation, so studies using other terminology may have been missed.
• No review protocol was registered in advance, and the corpus concentrates on Asian contexts and adult learners.

Connected Concepts

• Self-Efficacy
• Motivation
• Anxiety and Stress
• Generative AI
• Conversational AI
• Theories and Frameworks
• Meta-Analysis and Systematic Review
• Self-Report Measures
• Multilingual Learning

Connected Articles

• Associations Between Generative AI–Based Pronunciation Feedback and Willingness to Communicate in English: The Mediating Role of English Pronunciation Self-Efficacy — Pronunciation Feedback and Willingness to Communicate in English
• A Systematic Review of Emerging Technology Applications for Teaching English as a Foreign Language Across Different Educational Levels — Systematic Review of Emerging Technology for EFL Teaching
• Acceptance of AI-Gamified Adaptive Learning in EFL: Bifactor Evidence for Essential Unidimensionality and the Limits of Subscale Scoring — Acceptance of AI-Gamified Adaptive Learning in EFL

Citation

Zhang, S., Zou, B., Liu, Y., Chen, Z., & Yan, R. (2026). Impacts of artificial intelligence on willingness to communicate and its key antecedents in EFL contexts: a systematic review (2015–2026).