Empowerment over enforcement: unpacking the psychological drivers of AI-assisted deep revision in EFL writing

Synthesis: Using survey data from 327 Chinese EFL undergraduates and partial least squares structural equation modeling (PLS-SEM), this study tests whether AI prompting literacy or top-down external mandates better drive deep-revision engagement in AI-assisted academic writing. It finds that prompting literacy predicts deep revision through the parallel psychological mediators of perceived competence, intrinsic motivation, and psychological safety — while external mandates show no direct effect — reframing AI adoption from enforcement to internal need satisfaction.

Key Findings

• Prompting literacy is the engine of deep revision. Students with stronger AI prompting literacy reported markedly higher perceived competence, intrinsic motivation, and psychological safety — the three psychological needs that in turn sustain deep revision. Prompting ability functions as a personal resource that empowers learners rather than a mere technical trick.
• External mandates do nothing on their own. The direct link between being required to use AI and engaging deeply with one's text was negligible and statistically non-significant. Compulsory, compliance-driven use may secure surface obedience, but it does not elicit the cognitively demanding work of restructuring a draft.
• The three psychological needs together transmit the effect. Prompting literacy predicts deep revision by raising learners' sense of competence, their intrinsic interest in writing, and their psychological safety in an AI interaction space — each pathway independently significant, confirming a full parallel-mediation model.
• Intrinsic motivation is the single strongest driver. Of the three psychological mediators, intrinsic motivation most strongly predicted deep revision engagement; perceived competence and psychological safety contributed but more weakly. Enjoying the process matters most.
• The model explains a substantial share of the variance — roughly half of the variation in deep revision engagement — suggesting the psychological account is meaningful, not marginal.
• Empowerment beats enforcement as a policy frame. Since mandates alone failed to elicit deep cognitive engagement, cultivating prompting capability and need satisfaction offers a more promising route than monitoring, compliance metrics, or punitive rules.

Study Design & Method

The study used a cross-sectional quantitative survey of 327 Chinese EFL undergraduates recruited from two higher education institutions (an application-oriented institute of technology and a regional normal university). Questionnaires were screened for attention-check failures, implausibly fast completions, and patterned responding, and all constructs were measured with established self-report scales adapted to the AI-writing context — prompting literacy, deep revision engagement, and the three self-determination-based psychological needs — alongside a newly developed three-item scale for external mandate, piloted on a separate group of students.

Data were analyzed with partial least squares structural equation modeling (PLS-SEM), chosen for its suitability to an exploratory, complex parallel-mediation model and non-normal data, with significance assessed through bootstrap resampling. Because the design is cross-sectional and single-source, the authors interpret the mediation paths as model-dependent indirect associations rather than causal sequences.

What this means for practice

• Instructors. Teach prompting as a cognitive support strategy across drafting, feedback, and rewriting — the paper's Collaborative Human-AI Revision Design uses constraint-based prompts and criteria-specific diagnostic prompts that position AI as a formative peer reviewer rather than an authoritative oracle.
• Administrators. Pair necessary AI Governance with AI-prompting workshops and autonomy-supportive environments instead of monitoring-centric usage metrics or AI Detection software, because external mandates had no significant direct effect on deep revision (model R² = 0.500).
• Instructors. Move Assessment toward process and Metacognition: prompt-reflection portfolios that reward productive failure sustain the intrinsic motivation that was the strongest single predictor of deep revision, whereas summative rubrics that heavily penalize linguistic inaccuracy risk nudging students toward algorithmic dependence.
• Administrators. Evaluate students' iterative dialogue with AI as part of the graded work, treating human-AI collaboration as a core literacy rather than scoring only the final product.

Limitations

• Cross-sectional, single-source self-report data make the findings susceptible to common method bias; all mediation paths are model-dependent indirect associations, not causal or chronological links.
• No demographic moderators were modeled — the study does not test moderating paths or multi-group variance (e.g., gender differences) within the structural model.
• Quantitative scales may miss dynamic nuance — deep revision is a complex, iterative cognitive process, and self-report instruments cannot capture the full dynamics of human–computer interaction.
• Single-context generalization — findings rest on Chinese EFL freshmen and sophomores, limiting cross-cultural applicability.

Connected Concepts

• Cognitive Offloading — APL lets learners offload lower-level linguistic operations, freeing working memory for higher-order text reconstruction (Cognitive Load Theory).
• Self-Efficacy — Perceived Competence mediates the effect of prompting literacy on deep revision engagement.
• Motivation — Intrinsic Motivation, grounded in Self-Determination Theory, is the strongest direct driver of deep revision.
• AI Literacy — AI Prompting Literacy is conceptualized as the core psychological empowerment resource.
• Self-Regulated Learning — Deep Revision Engagement reflects iterative, self-regulated text reconstruction rather than surface-level editing.
• Feedback — AI is reframed as a formative, zero-judgment feedback interlocutor that supports psychological safety.
• Writing — EFL academic writing is the empirical domain, framed as a cognitively demanding iterative process.
• Language Learning — Study population of EFL learners in Chinese higher education.

Connected Articles

• Student Motivation and Need Satisfaction in GenAI-Supported Classrooms: A Self-Determination Theory Perspective — Shared SDT framing of psychological need satisfaction in generative-AI learning contexts.
• Are Students Dependent on AI in Writing Courses? Analyzing Factors Influencing Dependence on Generative AI Through the I-PACE Model — The I-PACE model of AI dependence in academic writing, mirroring the "algorithmic dependence" risk examined here.
• Understanding Student Dependency on AI: The Role of AI Literacy, Academic Self-Efficacy, and Resource Management Strategies — Related examination of AI literacy and self-efficacy in student AI use.
• Cognitive Offloading in Student–AI Collaboration: A Longitudinal Analysis of Prompting Strategies — Cognitive offloading in prompting, the same CLT mechanism underlying APL's effect.
• Penny: Transition Network Analysis of Learner-Chatbot Interactions in Scaffolded EFL Writing — Empirical work on EFL writing, the shared target population and construct of deep engagement.

Citation

Li, H., & Zhang, W. (2026). Empowerment over enforcement: unpacking the psychological drivers of AI-assisted deep revision in EFL writing. Frontiers in Psychology, 17, 1871022.