<![CDATA[Reading literacy is a fundamental competency determining academic success, yet Indonesian elementary students rank 71st out of 81 countries in PISA 2022. This study examined the integration of Large Language Models (LLMs) as adaptive reading assistants to enhance elementary students' reading literacy through three objectives: (1) developing an LLM-based adaptive reading assistant prototype; (2) measuring its effectiveness; and (3) exploring students' learning experiences. An explanatory sequential mixed-method design with a quasi-experimental nonequivalent control group involved 72 fifth-grade students from two public elementary schools in Bekasi (experimental n=36; control n=36). Data were collected through a PIRLS-adapted reading literacy test, classroom observations, and semi-structured interviews. Quantitative data were analyzed using ANCOVA with partial eta squared (η²ₚ) as the effect size estimator; qualitative data underwent reflexive thematic analysis. Results: (1) The prototype successfully delivered adaptive questioning, scaffolding, and Socratic dialogue; (2) the experimental group exhibited significantly higher gains (M=23.45) than the control group (M=9.82), F(1,69)=50.21, p<0.001, η²ₚ=0.42, 95% CI [10.97, 16.29], Cohen's d=1.42; and (3) students reported increased engagement, confidence, and self-regulated learning. This study is the first to operationalise cognitive apprenticeship principles in an Indonesian-language LLM prompt designed for elementary readers, combining a quasi-experimental evaluation with ethical safeguards for minors. LLM integration is an effective pedagogical strategy when complemented by teacher mediation.]]>
