2026/03/17 by Jonathan Ben-Menachem · 1 voice
Social Sciences · #Coding (social sciences) #Computational and Text Analysis Methods #Data Analysis and Archiving #Data collection #Incentive #Qualitative Research Methods and Ethics #Qualitative property #Qualitative research #Robustness (evolution)
paper · doi:10.31235/osf.io/gj89u_v1
openalex publication_date 2026/03/17 · openalex created_date 2026/03/18 · openalex updated_date 2026/07/14
Researchers are now considering how to incorporate large language models (LLMs) into qualitative data collection and analysis—domains where the researcher's evolving interpretive judgment cannot be separated from the method. This article considers various, increasingly ambitious applications of LLMs, distinguishing amplification of human analytical capacity from substitution. For analysis of human-collected data, I argue that the discipline lacks epistemic frameworks to evaluate what LLM-assisted coding means—whether it constitutes an informal thinking tool, a robustness check, or a source of confirmation bias. Turning to data collection, I suggest that LLM-administered “interviewing” sacrifices the iterative linkage between data collection and theory development that gives qualitative methods their distinctive inferential leverage, producing output closer to adaptive surveying than to fieldwork. I conclude by addressing institutional incentives that make these tools structurally tempting and arguing that professional evaluation standards should account for the epistemic costs of LLM-accelerated productivity norms on fieldwork-intensive research.