Memory allocation, parameter passing, and dynamic data structures in C++ are some large conceptual challenges high school students face. Because memory management is such a fundamental feature of C++, students generally develop incomplete or inaccurate mental models about what to do with stack and heap memory, semantics of references, and how data structures (like stacks, queues, and linked lists) that are dynamically allocated behave. Traditional methods of instruction and assessment are still concerned with syntactic correctness and problem-solving outcomes and miss out on deeper conceptual misconceptions. These beliefs and misconceptions can be diagnosed and corrected in 11th-grade learning of programming with the use of an AI-aided educational method proposed in this paper. Pre-test/post-test design study applied a conceptual questionnaire, aimed at generating students’ mental models, beyond basic syntactic knowledge. Each student’s written explanation and code interpretation were then analyzed in the context of an AI-based classification system based on an explicit ontology of common misconceptions. Facing the detected conceptual errors, the system responded with tailored feedback and problem exercises based on the specific needs of individual students. Results reveal a significant decrease in recurring misconceptions following the AI-assisted intervention, especially among students with weaker conceptual understanding after initial stages of the lesson.