Students’ Interaction Patterns with ChatGPT in Solving Image-Based Mathematics Problems

Abstract

Generative artificial intelligence (AI) tools such as ChatGPT are increasingly used by students to obtain explanations and solutions to mathematics problems. However, access to answers does not necessarily mean that students examine the underlying reasoning. This study aimed to describe students’ interaction patterns with ChatGPT when solving image-based mathematics problems, variations in these patterns across two problems, and how students responded to, sought clarification of, or challenged ChatGPT-generated solutions. A qualitative descriptive design was employed. The participants were Grade 9 students of one intact class (total sampling), and the records of 31 students who worked on two image-based problems, concerning the area of a trapezoidal footpath and the painted area of a photo frame, were analyzed. The primary data consisted of student–ChatGPT conversation transcripts exported from shared conversation links. The transcripts were analyzed descriptively and qualitatively using six predefined interaction patterns: requesting a solution and then ending the interaction; seeking clarification; requesting a re-explanation; checking an answer or testing a method; following a solution step by step; and connecting or redirecting the interaction. The unit of analysis was the student–problem episode, and overlapping codes were permitted. The findings indicate that students did more than request answers. Some asked for the rationale behind a solution step, proposed strategies for confirmation, requested further explanations, responded to practice questions posed by ChatGPT, and made connections between the two problems. Two students questioned the complexity or necessity of ChatGPT’s chosen method. Differences between the problems were primarily reflected in the content of students’ questions. These findings are descriptive and are not intended to establish the prevalence of interaction patterns, learning outcomes, or causal relationships. The findings suggest a need for task design and teacher guidance that encourage students to verify and evaluate AI-generated solutions.