How students engage with LLM-based GenAI to develop computational thinking: A systematic review

This systematic literature review examines how students engage with large language model-based generative artificial intelligence (GenAI) in developing computational thinking (CT). Following PRISMA 2020 procedures, 14 of 212 peer-reviewed studies published between 2021 and 2026 were synthesized to investigate how GenAI functions along a continuum from problem-solving tool to thinking partner. The review identifies four student-AI interaction patterns: surface, scaffolded, achiever-linked, and dialogic. Students with stronger CT proficiency and AI literacy engaged GenAI dialogically through critique, verification, and reflective prompting, whereas lower-proficiency learners relied on answer-oriented interactions linked to substitutive offloading and weaker problem-solving. GenAI supported algorithmic thinking, creativity, and reflection when students retained task agency, but risked over-reliance otherwise. Most studies positioned GenAI at a cognitive tool mid-point, yet sustained CT development depended on designs promoting epistemic agency and reflective dialogue. The review contributes a continuum framework and seven pedagogical safeguards that shift AI use toward thinking partnership.