Enhancing L2 writing skills at B2 level through adaptive artificial intelligence: Personalized learning approaches and their impact on L2 writing skills

This study investigated the impact of AI-assisted EFL writing instruction on attitudes, perceptions, problems and writing skills among tertiary-level ELT students, as well as the strategic utilization of AI by learners within a full year-long writing program. A concurrent embedded mixed-methods design was executed across two semesters alongwith a quasi-experimental pre/post design with 56 freshman ELT students. Quantitative data comprised pre–post writing proficiency scores for both the experimental (AI-integrated) and control groups, three-point (pre-mid-post) assessments of writing attitudes, perceptions of AI-assisted writing and identified problems. The qualitative data comprised weekly reflecting diaries and semi-structured interviews. Triangulation was employed to synthesize item-level patterns with thematic insights. Results indicated that writing proficiency enhanced in both groups, but the experimental group exhibited markedly larger gains (d = 3.93 vs. 0.68) and a significant post-test advantage (d = 0.93). Qualitative findings validated improvements in linguistic accuracy, structural organization, genre awareness, metacognitive monitoring, and strategic revision. In addition, perceptions of AI-assisted writing were consistently favorable, showing a modest rise mid-semester followed by a return to baseline, yet this fluctuation was not statistically significant. This insignificant slight difference demonstrated a "honeymoon effect" or novelty impact, wherein the initial enthusiasm for gamified revision, noticeable score enhancements, and immediate AI involvement temporarily increased perceptions. As AI integration gradually became typical, attitudes stabilized through a routinization effect, yet remained consistently positive. Students perceived AI as a complementary collaborator instead of a substitute for teachers and cultivated a critical awareness of its constraints. Moreover, perceived problems remained modest and consistent throughout two semesters, with only cost-related concerns experiencing a statistically significant decline. Technical challenges, conflicting feedback, and ethical dilemmas fostered AI literacy, critical judgment, and responsible authorship instead of discouragement. Furthermore, attitudinal results indicated minor yet statistically significant reductions in enjoyment-related items; however, qualitative data demonstrated a favorable developmental transition: Students progressed from fear and uncertainty to confidence, comfort, and self-efficacy as AI reduced anxiety and clarified expectations. An essential explanatory pattern emerged: As students participated in more iterative, multi-draft writing tasks utilizing AI technologies, they developed an increased awareness of the cognitive demands associated with academic writing. This higher level of awareness, along with continuous revision cycles, resulted in cognitive fatigue and caused students to perceive writing not as a light or enjoyable activity but a demanding academically important skill, accounting for the observed decrease in fun despite enhanced proficiency. Overall, AI functioned as a pedagogically beneficial, teacher-facilitated scaffold that enhanced traditional process-based writing instruction and fostered cognitive, linguistic, and metacognitive growth in L2 writing. In light of the findings, several pedagogical implications were provided to the teachers and a comprehensive "AI Integrated EFL Writing Syllabus Guide" was prepared for EFL writing teachers to be a "manual" or "handbook" to follow.