Large Language Models as Programming Tutors in Computer Science Education: A Systematic Review of Applications, Learning Effectiveness, Error Patterns, and Academic Integrity

Abstract

This review synthesizes empirical evidence on how large language models (LLMs) are used as programming tutors across five key applications: code explanation, debugging, formative feedback, exercise/test generation, and assessment. FollowingPRISMA2020guidance, studies from 2020–2025 were screened for relevance to programming education, with outcomes on learning effectiveness, error patterns, and academic integrity synthesized via narrative and thematic methods. The mapping of recent studies indicates rapid up take of LLMs, mixed but promising learning outcomes, recurring failure modes in logic and specification adherence, and emerging academic integrity risks alongside mitigation practices. Implications are provided for course design in Object-Oriented Programming(OOP)contexts, assessment practices, and departmental policy, with identified gaps and recommendations for standardized evaluation.