Many reading literacy programs in primary schools raise student scores, yet few explain why those gains occur. This study examines how the FIVES model (Fact, Inference, Vocabulary, Experiences, Summary) supported by artificial intelligence (AI) works for Grade 5 students in twelve primary schools across five Indonesian provinces. A convergent mixed methods design was used. Quantitative data came from a reading literacy pretest and posttest (N=172), a self-regulated learning questionnaire (N=277), and a reflective learning instrument (N=208). Qualitative data came from classroom observations of seven teachers, interviews with nine teachers, and interviews with 27 students. Both strands were analysed separately and then brought together in a joint display. Mean literacy scores rose from 54.11 to 81.31 (N-Gain 0.584, Cohen's d 2.46), and the spread of scores between students narrowed. The integration produced five findings. AI helped students see the reasoning process at the Inference stage. The FIVES structure offered different entry points for students of different abilities. The way teachers guided AI use mattered more than the availability of AI itself. Reflective skills proved learnable for all students. AI also offered a safe space for students who had been reluctant to ask questions. These findings suggest that strengthening teachers' pedagogical capacity to guide AI use should be a priority in technology-supported literacy programs in primary schools.