How does generative AI affect self-regulated learning in learners with varying knowledge levels? The role of chatbot design and support timing

Generative AI (GenAI) chatbots can provide on-demand support, but their educational value hinges on how learners regulate its use. Guided by self-regulated learning (SRL) theory, this study examined how two design dimensions of GenAI support shaped students’ learning outcomes, interaction behaviors, and subjective perceptions in physics learning tasks. The sample included 107 students in a 2 × 2 design crossing support timing (immediate vs. delayed) and chatbot response design (open vs. scaffolded), with prior knowledge modeled as a continuous variable. Interaction logs during GenAI use were coded according to SRL phases and analyzed using lag sequential analysis. Results showed that delayed support led to higher learning outcomes than immediate support, whereas chatbot response design had no significant effect on learning outcomes. At the process level, immediate support was associated with more direct transitions from initial prompting to answer adoption, whereas delayed support created more opportunities for questioning, evaluation, and revision. Higher prior knowledge was associated with stronger perceived interactivity and, in the delayed support and open response condition, higher perceived ease of use. These findings suggest that effective GenAI learning support may depend on how support timing and response structure engage learners in self-regulated processing of content generated by GenAI.