Generative AI in Vietnamese EFL Self-Study: Task Patterns, Usefulness, and Student Judgement

Generative AI has entered student self-study rapidly, yet its use remains uneven across language-learning tasks. This explanatory mixed-methods study investigated how 200 English-major students at a public language university in Central Vietnam used Gemini in out-of-class English learning and how they evaluated its usefulness. Survey data were collected to examine frequency, task distribution, and perceived usefulness, followed by interviews with 10 students to explain patterns of use. Drawing on self-regulated learning (SRL), task–technology fit, and evaluative judgement as interpretive lenses, the findings revealed regular but bounded use rather than broad dependence. Most students used Gemini several times per week in short or moderate sessions, treating it as a support resource rather than a complete learning environment. Use was strongest in text-based tasks such as drafting emails, completing assignments, checking grammar, clarifying vocabulary, and paraphrasing, where learners could inspect and revise generated output. Use was weaker in speaking, dialogue comprehension, and audio-based activities that required natural interaction and real-time processing. Students valued Gemini for its speed, accessibility, and convenience, but usefulness was judged selectively rather than uniformly. Participants frequently evaluated tone, contextual fit, and cultural appropriateness before deciding whether to adopt AI-generated responses. From an SRL perspective, Gemini appeared to support specific episodes of planning, monitoring, and revision rather than full self-regulated learning. The study argues that Gemini functions primarily as a bounded support tool within EFL self-study, offering strongest value in tasks where learners retain evaluative control over generated output.