This study examines authentic student interactions using the publicly available CoMTA dataset from Khan Academy's AI-driven tutor, Khanmigo.

The analysis explores three core dimensions:
(1) The volume and significance of student-generated questions, demonstrating that students interacting with Khanmigo have substantially more opportunities to ask questions compared to traditional classroom or individual learning contexts;
(2) The effectiveness of Khanmigo's dialogic approach, characterized by prompting independent reasoning rather than providing immediate answers, thereby actively supporting students' cognitive engagement and discovery processes; and
(3) An investigation into the progression of question types within individual tutoring sessions to determine if and how student inquiries transition from requests for direct answers or procedures toward questions that reflect deeper confirmation of understanding, reasoning, or exploratory behavior.

Insights from this analysis highlight the role of conversational AI tutors like Khanmigo as facilitators of critical thinking and self-directed learning, offering valuable implications for instructional design and future educational technologies.

Keywords: AI tutoring, student inquiry, cognitive taxonomy, question classification, CoMTA dataset, large language models, educational technology, dialogic learning.