As being more prevalent in educational settings, understanding the impact of artificial intelligence tools on student behaviors and interactions has become crucial. In this regard, this study investigates the dynamic interactions between students and ChatGPT in programming learning, focusing on how different instructional interventions influence their learning and AI-interaction. Conducted over three sessions, students were allowed to use ChatGPT to complete programming tasks. The first session had no guidance, the second included hands-on training in prompt writing and effective ChatGPT use, and the third provided a lab guide with sample prompts. After each session, students took a post-test on the activity's subject. Analyzing students' prompting behaviors, five AI interaction profiles were identified: AI-Reliant Code Generators, AI-Reliant Code Generator & Refiners, AI-Collaborative Coders, AI-Assisted Code Refiners, and AI-Independent Coders. These profiles were examined to understand their evolution across interventions and their relationship with students' learning performance. Findings revealed significant changes in profile distribution across interventions, and a notable difference between students' post-test scores and their AI interaction profiles. Besides, training in prompting skills and effective use of AI significantly impacted students' interactions with AI. These insights can contribute to the knowledge of integrating generative AI tools in education, highlighting how AI can enhance teaching practices. Understanding student-AI interaction dynamics can allow educators to tailor instructional strategies for optimal learning. This study also underscores the importance of guidance on effective AI use and prompting skills, which can lead students to use AI more meaningfully for their learning.