ChatGPT in Introductory Programming: Counterbalanced Evaluation of Code Quality, Conceptual Learning, and Student Perceptions

Background: Large language models (LLMs) such as ChatGPT are increasingly used in introductory programming courses to provide real-time code generation, debugging, and explanations. While these tools can boost productivity and code quality, concerns remain about over-reliance and potential impacts on conceptual learning. Objective: To investigate how ChatGPT access affects code quality, conceptual understanding, task completion times, and student perceptions in a CS1 course. Methods: We conducted a counterbalanced, quasi-experimental study in which students alternated between ChatGPT and non-ChatGPT conditions across two programming assignments in C (functions and structures). We evaluated their code submissions using multidimensional rubrics, conceptual post-surveys, and task completion time. Results: Students who had access to ChatGPT produced significantly higher rubric scores for code quality and completed tasks in less time compared to those without access. However, gains in conceptual understanding were mixed, lower for the functions topic but higher for the structures topic. Students reported positive experiences with ChatGPT, citing its value for debugging and practice, while expressing concerns about accuracy and long-term skill development. Conclusions: ChatGPT can enhance code quality and efficiency for novice programmers, but may not uniformly improve conceptual understanding. Structured integration and complementary instructional strategies are recommended to foster independent problem-solving skills.