EFL learners' emotional engagement in GenAI-assisted writing: a longitudinal multiple-case study

Synthesis: Wu, Yang and Lin (2026) traced how three Chinese EFL learners' emotional engagement with GenAI-assisted writing moved across three tasks in a 14-week first-year medical English course, asking which individual and environmental factors shaped it. The focal students came from 14 volunteers drawn from a survey of 241 students in seven classes, followed through three rounds of stimulated-recall interviews (7 h and 5 min in total), reflective reports, chatlogs and drafts. The authors name three trajectories: positive-resilient for Leo, pragmatic-transformative for Grace, and vulnerable-contingent for Sarah, whose path ended in a breakdown that included leaving the classroom mid-task. With three participants, the study can show these trajectories are possible and describe what produced them in these three learners. It cannot estimate how common any is, and it supports no claim about EFL learners as a population.

Key Findings

1. Three focal cases came from 14 purposively sampled volunteers out of 241 students surveyed across seven classes, traced across an exposition, a group article and an argumentation over 14 weeks.
2. All three were 19-year-old Chinese L1 first-year medical students taught by the first author, differing at entry: Leo advanced (13 years of English, weekly GenAI use), Grace intermediate (most prior essays drafted with GenAI), Sarah beginner (9 years of English, Japanese as her exam subject).
3. Leo's positive-resilient path included real dips: shaken by the errors GenAI surfaced, restored by a 95 on his first draft, frustrated by the group task yet scoring 85 there and 89 on argumentation.
4. Grace's pragmatic-transformative path began with a 92 on her first draft alongside insecurity about her English, moved through peer-mediated group writing she found helpful (80), and turned on finding grammatical errors in GenAI's own text, after which she verified sources and reached 90.
5. Sarah's vulnerable-contingent path began at 76 with a note explaining her Japanese-language background that went unread because only paper drafts were graded, rose to 84 in the group task, then broke down: she left the classroom during argumentation and submitted after class for 78.
6. Identical features cut opposite ways: group writing lowered anxiety for Grace and Sarah but frustrated Leo, and English-medium workshops served Leo and Grace and blocked Sarah.

Three students out of two hundred and forty-one

After the survey of 241 students, 14 were chosen for willingness, proficiency band and prior GenAI experience, and the three focal cases were picked afterward for varied proficiency, gender, GenAI history and, explicitly, data richness. Proficiency bands came from previous-semester course scores (advanced 85 to 100, intermediate 70 to 84, beginner below 70), a local grouping rather than a standardized benchmark. Data included stimulated-recall interviews after each task, reflective reports, chatlogs, drafts and informal chats, coded thematically by both authors, the second partly offsetting the first author's knowledge of her own students. Transcripts were machine-translated with DeepSeek-V3, checked by both authors, and back-translated in sample.

The three trajectories, as the authors name them

Positive-resilient describes Leo: sustained positive affect with temporary fluctuation, and his dips are not disengagement because his resources let him reframe frustration as information. Pragmatic-transformative describes Grace: practical disengagement giving way to critical engagement. Her first stage was efficiency-driven, prompting DeepSeek for definitions and examples because she was "too lazy to think," which produced relaxation and few original ideas; by the third she was verifying whether GenAI's cited example sources held up. Vulnerable-contingent describes Sarah: moderate disengagement, tentative peer-mediated improvement, then breakdown and intensified disengagement, her second-stage engagement contingent on support that never restructured her underlying beliefs.

Disengagement is defined only where negative descriptors predominated in frequency, intensity and persistence rather than being recovered from, which lets Leo's frustration count as engagement and Sarah's as its opposite.

Why the same course felt different to each

The sociocultural account declines to rank individual above environmental causes. Proficiency and GenAI literacy do most of the work: proficiency decided who could follow the English-medium workshops, while GenAI literacy moved Leo from experienced to strategic user, Grace from unguided heavy user to critical one, and left Sarah rudimentary. Environmental features entered at task, classroom and policy level. The sharpest case is failed mediation: Sarah's note about her Japanese-language background went unread because only paper drafts were graded. The authors read this through perezhivanie, the prism through which a setting is refracted, and argue that mediation must be calibrated rather than merely available.

What three cases cannot show

A multiple-case design can show that divergent trajectories occur inside one course and specify mechanisms worth testing. It cannot do what the paper's language sometimes approaches: characterize GenAI's affective impact as heterogeneous in general. Three students chosen for maximum difference are, by construction, maximally different, so selection partly guarantees the heterogeneity reported. Prevalence is not estimated, and neither is the direction of group differences, since genre, format and position in the semester changed together across the three tasks, a confound the authors name.

What this means for practice

For instructors, the practical task is routing support by profile, not choosing a tool. Advanced learners need agency-preserving challenge rather than more Scaffolding. Efficiency-driven learners need tasks that make their GenAI use visible and accountable, since the multi-step requirements were what turned Grace's annoyance into scrutiny. Vulnerable learners need affective support plus foundational language and GenAI literacy, delivered in a language they can process: Sarah's exclusion began with workshops she could not follow. Teacher attention is what no tool replaces; the costliest failure here was an unread note.

Limitations

• Three participants chosen for difference cannot support any statement about how often a trajectory occurs; the authors call the framework adaptation tentative and the interpretation provisional.
• Genre, format and time in the semester varied together, so change attributed to development cannot be separated from task effects.
• All data come from one course, with the first author as instructor of the same students; interviews were scheduled before grades were released, but the dual role is not removed by scheduling.

Citation

Wu, K., Yang, M., & Lin, C. (2026). Engagement or disengagement: understanding the emotional dynamics of EFL learners in GenAI-assisted writing from the sociocultural perspective. Frontiers in Psychology, 17, 1934782.