Using Generative AI for Qualitative Coding

Abstract: Researchers at the University of New Mexico used ChatGPT 4 to replicate the coding and analysis process of a qualitative research study investigating the lived experiences of graduate students. The study’s purpose was to explore the trustworthiness and credibility of generative AI for the coding process in qualitative research. The researchers compared the first- and second-level codes and the thematic framework produced by a human research team with the first- and second-level codes and thematic framework produced by ChatGPT 4. They conclude that ChatGPT was effective for first-level open and descriptive coding and that researchers who use ChatGPT for their first-level coding will save a substantial amount of time and can devote more attention to second-level coding. A compressed timeline could not only benefit researchers but also speed up the adoption of data-informed improvements, yielding palpable benefits for the communities that libraries serve. The researchers also conclude that ChatGPT’s inherent limitations mean that it cannot serve as the primary second-level coder. They recommend using ChatGPT as a nonhuman collaborator, with options to use it in the capacity of researcher triangulation, intercoder reliability, peer debriefing, or reduction of bias. If researchers use ChatGPT for second-level coding, they will still need to allocate sufficient time to gaining intimate knowledge of the transcripts in order to produce more trustworthy and credible results.