<p>Unlike static assessments, dynamic assessment evaluates learning potential by providing prompts and scaffolds during interactive cognitive tasks that blend instruction and assessment. Designing prompts and scaffolds for dynamic assessment that align with both the content of reading materials and students’ language proficiency is challenging. Generative artificial intelligence (AI) tools offer new solutions to this challenge. Leveraging databases, text processing, and machine learning, generative AI can select texts and design questions based on students’ proficiency, as well as create prompts that guide the framework of dynamic assessment. In this study, we trained a generative AI tool to design two sets of dynamic assessments focused on Chinese lexical inference and reading comprehension for second-grade Chinese-speaking elementary school students. Students were required to complete five-option multiple-choice questions for both dynamic assessments on a computer. Each incorrect choice triggered a corresponding prompt, ranging from implicit to explicit. Mediated scores and learning potential scores (LPS) on both assessments accounting for the weight of each mediating prompt received by test-takers were calculated and compared with their performance on reading-related subskills (phonological awareness and word reading). Regression analysis revealed that after controlling for age and non-verbal intelligence, mediated scores on the reading comprehension statistically significantly predicted performance in word reading. Additionally, mediated scores on lexical inference were found to predict both word reading and phonological awareness. These findings confirm the alignment between AI-assisted dynamic assessment and test-takers’ language proficiency, as measured by both implicit metalinguistic awareness and explicit word reading ability.</p>