Abstract
This study utilizes GPT-4o, an LLM, to interpret semi-structured interview narratives into structured numerical indicators of wellbeing. Key findings of this study are: firstly, different LLM instances produce consistent numerical wellbeing results. Secondly, we observe close alignment of these quantitative wellbeing results with the qualitative wellbeing interpretations. Thirdly, the strong correlation between the LLM’s PHQ-9 and GAD-7 scores suggests that the model treats these constructs as a single general wellbeing measure when assessing workplace narratives. The study is based on 18 interviews (21 hours of speech) conducted between 217 and 221 with four IT professionals. Our study presents the LLM as a methodological tool for Information Systems (IS) researchers to scale mixed-method analysis. The study contributes (1) a demonstration of how an LLM can be applied to workplace narratives and (2) a validation and limitations when utilizing an LLM in producing numerical results for wellbeing questionnaires.
Recommended Citation
Hekkala, Riitta and Nyberg, Riko, "From Narratives to Numbers: Evaluating a Large Language
Model (LLM) for Transforming Workplace Interview Narratives
into Numerical Wellbeing Indicators" (2025). ACIS 2025 Proceedings. 240.
https://aisel.aisnet.org/acis2025/240