Qualitative content analysis of learners’ conceptions is due to large datasets time-consuming. This study examined the potential of a Large Language Model (LLM) to support and accelerate qualitative content analysis without compromising validity. Written responses from 180 bachelor students at two German universities to four open-ended questions on microplastics were analysed. ChatGPT was used to inductively develop categories and to assign responses. Two human expert coders conducted the same procedures for comparison purposes. The inter-rater reliability was calculated using Cohen’s kappa and two independent ChatGPT runs were performed to test consistency. The categorization system generated by ChatGPT largely corresponded to the human-developed system. The two ChatGPT runs showed highly consistent classifications, with inter-rater reliabilities of up to κ = 0.96. That exceeded both intra- and inter-rater agreement of the human coders ( κ = 0.45–0.90). Overall, our findings suggest that LLMs can support valid and reliable qualitative content analysis while substantially reducing analysis time. In this respect, the use of LLMs may represent a methodological game-changer for qualitative research, making such approaches more efficient and accessible, also for classroom teachers.
Braune et al. (2026) studied this question.