AI-Assisted Writing, Reliance, and Learner Autonomy: An Exploratory Mixed-Methods Study of Indian ESL University Students

Artificial intelligence writing tools are increasingly embedded in how university students draft and revise academic English, raising questions about their relationship to learner autonomy. This exploratory mixed-methods study surveyed 71 Indian ESL university students using 20 Likert-scale items across four conceptual domains and five open-ended questions, addressing how learners perceive AI's benefits and drawbacks, how these perceptions structure, how learners describe dependency and confidence in their own words, and what they suggest for responsible classroom use. Minimum-residual exploratory factor analysis with oblimin rotation, guided by common-factor parallel analysis, favoured a four-factor solution: Confidence/Proficiency, Writing Assistance/Reliance, Ethics/Awareness, and Creativity/Originality Concern. Confidence, Assistance, and Ethics were positively associated at the respondent level (Spearman ρ = .32–.47), while Concern was largely unrelated to the other three. Three dependency-worded items did not behave as interchangeable indicators: over-reliance aligned with Writing Assistance/Reliance, perceived dependency aligned with Creativity/Originality Concern, and difficulty writing without AI remained unstable across specifications. Qualitative findings echoed this heterogeneity: benefits named were predominantly organisational and mechanical, challenges centred on loss of personal voice and over-reliance, and learners combined benefit, ethical awareness, confidence, and concern differently across respondents, rather than converging on one pro- or anti-AI orientation. Interpreted through Benson's (2011) dimensions-of-control framework, Writing Assistance/Reliance bears on learning management but cannot distinguish learner-managed support from delegation to AI. Concern and Ethics/Awareness relate provisionally to different strands of learning content, and Confidence/Proficiency falls outside the framework entirely. When asked directly, learners most often favoured AI as a supplementary aid rather than a substitute for independent writing, alongside a preference for teacher-guided over unrestricted classroom integration. The findings challenge unitary accounts of AI dependency and suggest educators should distinguish behavioural assistance-seeking from concerns about creativity, ownership, and self-reliance when designing that guidance.