AI-powered learning: exploring secondary school students’ use of ChatGPT for solving mathematical modelling problems

Abstract

Abstract The rapid emergence of large language models has raised important questions about their potential to influence learning and problem-solving in real-world contexts. Mathematical modelling is particularly relevant in this regard, as it links abstract mathematical concepts to real-life situations and plays a key role in mathematics curricula. However, few studies have examined how school students use these tools in practice. This study investigates how secondary school students use ChatGPT to solve mathematical modelling problems, a core element of mathematics education. In this exploratory study, ten tenth-graders worked in pairs during a two-hour session on two modelling tasks using ChatGPT. Data sources included written work, chat logs, and semi-structured interviews. Findings show that students relied almost exclusively on zero-shot prompting; few-shot and chain-of-thought strategies were absent, and output customisation was rare. Most prompts focused on early modelling phases (simplifying, structuring, mathematising), while validation and interpretation were largely missing. Prompts often lacked coherence, limiting deeper reasoning. Outcomes also differed by task, none of the groups solved the visual Taj Mahal problem; whereas the verbal E-Car problem was handled more successfully. ChatGPT supported assumption generation and reduced workload, but uncritical acceptance and limited verification indicated superficial engagement. This study extends research on pre-service teachers by providing classroom evidence from school students. Results suggest that while ChatGPT can lower barriers in early modelling, explicit training in prompt literacy and modelling competencies is essential for critical and meaningful use.

The authors' abstract, as published at the source. ZDM, 2026 · DOI ↗