["Journal of Computer Assisted Learning, Volume 42, Issue 5, October 2026. ", "\nABSTRACT\n\nBackground\nLearning by Teaching (LbT) is regarded as an effective generative learning approach. However, its effectiveness is often constrained by implementation conditions, as well as the cognitive level and motivation of students acting as learners, limiting its full potential. With the widespread application of generative artificial intelligence (GAI) in education, this approach has gained new opportunities for development. Nevertheless, existing studies have mainly focused on the “student questions–GAI answers” interaction, which essentially remains a form of passive learning and fails to promote deep knowledge construction.\n\n\nObjectives\nThis study introduces GAI into the LbT model, assigning it the role of a novice learner. Students shift from knowledge receivers to knowledge constructors, aiming to enhance their knowledge understanding, self‐efficacy and metacognitive strategies.\n\n\nMethods\nThe participants were 68 preservice teachers, including 33 in the experimental group and 35 in the control group. Two GAI roles were designed: generative artificial intelligence as a novice learner (GAI‐NL) and generative artificial intelligence as a teacher (GAI‐T). Students in the experimental group explained knowledge to GAI‐NL, while students in the control group asked questions to GAI‐T.\n\n\nResults\nThe results indicate that students in the experimental group showed significant improvement in the knowledge comprehension dimension. Their self‐efficacy was also significantly higher than that of the control group. Regarding metacognitive strategies, the experimental group scored significantly higher than the control group on the dimension of metacognitive knowledge and learning strategies; however, no significant difference was found between the groups in the ability to plan and monitor learning. In terms of attitudes toward the GAI role, the experimental group reported a higher mean score on perceived usefulness, whereas no significant difference was observed between the groups on perceived ease of use.\n\n\nConclusion\nThe GAI‐NL role operationalizes and extends the LbT model, creating a replicable and transferable learning environment. Through dialogue strategies such as Request for Explanation, Request for Examples, Verification Reasoning and Request for Supplementation, it guides students to deepen their understanding of knowledge. Students' self‐efficacy was strengthened, and their knowledge about their own learning improved, mainly supported by strategies such as Judgement, Opposing Perspective and Confirm Understanding. However, no significant improvement was observed in ability to plan and monitor learning, likely because the strategies focused more on knowledge presentation and understanding than on regulating the learning process. Future research should enhance GAI‐NL's support for learning planning and monitoring to comprehensively promote learners' metacognitive abilities.\n\n\nImplications\nBased on the role design of GAI as a novice learner, this study developed dialogue strategies to support the GAI‐NL role and empirically examined its effectiveness within the LbT model. The findings provide theoretical and practical evidence for reconstructing the roles of generative artificial intelligence in instructional contexts.\n\n"]