The advent of Generative Artificial Intelligence(GAI) represented by ChatGPT has challenged the traditional learning. How students learn through GAI tends to be an urgent problem to be explored in the current education and teaching reform. This research analyses the dialogues in class between undergraduate and GAI by coding discourse types, questioning levels, questioning strategies and students’ self report to explore the learning model of human-artificial intelligence collaboration. It is found that in student-GAI dialogues dominated by students, there are more single round conversations and less continuous discussions around a topic. The main types of students’ discourse are initial questioning, extended questioning and rephrasing questioning, while the evaluation and continuing instruction discourses are less. What’s more, students’cognitive level of questioning is low, focusing on knowledge level questioning and comprehension level questioning. The using of questioning strategies is unfamiliar and students seldom use role questioning, material questioning and scheme questioning. In addition, it is discovered that different task stages and different experience both lead to different conversation situation between students and GAI. With the development of task solving, there are more frequent and sustained dialogues, along with the deeper cognitive level and more proficient using of questioning strategies. Meanwhile, students with more experience in using GAI generate more dialogues with high cognitive level. In student-GAI dialogues representing different characteristics, though there are different opinions towards using GAI in class teaching, most of the students hold a positive attitude. In students’ perception, GAI has the advantage in generating responses, furnishing valuable information, handling various types of tasks and fostering the development of student abilities, thereby assisting students in learning. But at the same time, GAI faces challenges related to technical limitations, raising concerns about student development, learning assessment, and overall educational ecosystem. According to the findings of the research, our study provides effective suggestions for further introducing GAI into classroom teaching from three aspects: providing question guidance, enriching question scenes and strengthening reflection of GAI response.