<![CDATA[The use of generative artificial intelligence in teaching reading comprehension has expanded rapidly since late 2022. Despite growing interest among language educators, few studies have examined how teachers can systematically introduce AI tools into actual classroom reading instruction using iterative, reflective methods. This study aimed to answer two questions: first, whether structured AI integration across multiple instructional cycles improves EFL students' reading test scores, and second, how students' ways of interacting with AI change as they gain more experience. Forty second-year English majors at an Universitas Nusantara PGRI Kediri participated in three action research cycles over nine weeks. The researcher, who was also the classroom teacher, designed nine reading tasks where students used ChatGPT alongside academic passages. Data collection included pre-tests and post-tests measuring five reading sub-skills, screenshots of student-AI conversations (360 total), questionnaires administered after each cycle, semi-structured interviews with 12 students, and detailed classroom observation notes. Test scores rose from an average of 62.4% before the intervention to 79.25% at the end of Cycle 3. A repeated measures ANOVA confirmed that these improvements were statistically significant (p < .001) with a large effect size (partial η² = 0.676). The most notable gains appeared in two higher-order skills: synthesizing information across multiple text segments (+20.3 percentage points) and drawing inferences from implied content (+19.0 percentage points). An examination of the 360 ChatGPT interaction logs revealed a clear shift in how students used the tool. During Cycle 1, 62% of all prompts were either vocabulary definitions (34%) or translation requests (28%). By Cycle 3, these basic requests had dropped to 25% of total prompts, while more sophisticated prompts emerged: questioning an author's tone or bias (18%), asking for counterarguments (15%), and requesting feedback on the student's own understanding (12%). Interview data, analyzed using thematic analysis, produced four main themes. First, students consistently reported that ChatGPT helped them overcome comprehension barriers that would otherwise have stopped them from engaging with difficult academic texts. Second, participants described developing a more critical stance over time, learning to verify AI outputs and recognize potential inaccuracies. Third, students expressed ongoing tension between appreciating AI's helpfulness and worrying about becoming dependent on it. Fourth, those who worked in pairs during Cycles 2 and 3 reported that discussing AI responses with a partner reduced uncritical acceptance and deepened their understanding. When teachers integrate AI tools through sustained, reflection-driven classroom action research rather than one-time implementation, students can achieve measurable reading comprehension gains while simultaneously developing critical digital literacy. The findings suggest three practical principles for language educators: introduce basic prompting first, add collaborative structures to encourage peer discussion of AI outputs, and continuously adjust instruction based on observed student difficulties rather than following a fixed plan.]]>
Copyrights © 2026
