Research on the impact of generative AI interaction modes on the improvement of school-age children's early reading ability and college students' auxiliary intervention.

Background: Against the backdrop of digital empowerment, generative artificial intelligence has been widely applied in school-age children's reading education, and digital reading has become one of the primary types of environmental exposure for school-age children. However, school-age children face challenges such as insufficient age-appropriate content and uncontrollable physical and mental strain. There is an urgent demand for a public health intervention model that balances reading-related health benefits, prevents and controls digital exposure risks, avoids physical and mental health impairments, and adapts to school-age children from diverse socioeconomic backgrounds. Methods: A retrospective control study design was used in this study. A total of 100 cases of school-age children aged 6-12 were collected from the standardized intervention archive of our institution from September 2025 to December 2025. According to the intervention plan actually accepted in the archive records, they were divided into the collaborative intervention group and the AI-only reading group with 50 cases in each group. In the AI-only reading group, the recorded intervention is purely generative AI independent reading intervention (college students are only responsible for equipment operation and maintenance, and do not participate in reading guidance and emotional intervention), while in the collaborative intervention group, the recorded intervention is the structured auxiliary collaborative intervention of the whole process of generative AI voice multi-round interactive reading combined with standardized training for college students. The intervention period is 12 weeks, three times a week, and each time is 20-25 min. Retrospective analysis of longitudinal data at baseline, 6 weeks of intervention, 12 weeks of follow-up and 4 weeks of follow-up, with the comprehensive score of early reading as the only preset main outcome index, repeated measurement analysis of variance, independent sample Results: Retrospective archive data shows that, after intervention, the collaborative intervention group achieved significantly higher composite early reading scores, phonological awareness, vocabulary comprehension accuracy, reading attention duration and reading interest scores than the AI-only reading group ( Conclusion: The generative AI interaction mode paired with college student auxiliary support is associated with sustained improvement school-age children's core early reading ability via cumulative short-term follow-up effects. Meanwhile, this collaborative model enables effective management of digital environmental exposure risks and may reduce risks related to visual and psychological health among school-age children. The intergroup differences remained stable with reliable effect sizes. This intervention model can be may provide a reference for future implementation as a digital public health intervention targeting school-age children's early reading, and provides preliminary empirical evidence for the standardized application of AI-supported education for school-age children. This retrospective study only reflects the differences between groups of compound intervention programs in archived cases, and cannot establish strong causal inference.