Integrating Artificial Intelligence with Manual Coding of Qualitative Data: Opportunities and Challenges

Coding textual data is a vital but time-consuming part of qualitative data analysis. This article evaluates the integration of artificial intelligence (AI) tools with manual coding of participant interviews in a study examining how individual, interpersonal, and neighborhood factors affect overall health and cancer risk in the African American community. We conducted eighty-one semistructured, virtual, one-on-one interviews with stakeholders. The study team manually coded seventeen interviews (21 percent) to develop a coding framework of primary and secondary codes, which were then used to train the AI-assisted autocoding tool in NVivo. The tool autocoded the remaining sixty-four interviews (79 percent). The AI-assisted tool successfully extracted and matched secondary codes to their corresponding primary codes (e.g., barriers, facilitators, health care access, health outcomes). Notably, the AI-assisted tool achieved 100 percent accuracy in identifying participants’ neighborhoods and place names mentioned in the interviews, supporting spatial analysis of risk factors of cancer by location. The AI-assisted tool struggled, however, to distinguish nuanced secondary codes, such as interpreting safety contextually as a barrier or facilitator, requiring manual review and correction. Despite this limitation, the integration of AI with modest human intervention proved effective for efficient and scalable coding in large qualitative data sets. This integrated approach offers a promising strategy for accelerating coding while maintaining rigor in qualitative research.