ChatGPT as a qualitative research partner

Dr Lynette Pretorius

Contact details

Dr Lynette Pretorius is an award-winning educator and researcher specialising in doctoral education, academic identity, student wellbeing, AI literacy, research skills, and research methodologies.

Chris Pretorius

Contact details

Chris Pretorius is a doctoral candidate specialising in spiritual health and practice, with an interest in the intersections between technology and theology.

The rise of generative AI has sparked new conversations about its role in academic research. While generative AI tools like ChatGPT have proven effective for summarisation, pattern recognition, and text classification, their potential in deep, interpretive qualitative data analysis remains underexplored. In our recent study, we examine the integration of ChatGPT as an active collaborator in qualitative data analysis. Our findings highlight ChatGPT’s ability to streamline initial coding, enhance reflexivity and higher-order thinking, and support knowledge co-construction while emphasising the necessity of human oversight.

Our study marks an exciting step forward in the integration of generative AI into qualitative inquiry. By approaching generative AI as a partner rather than a passive tool, we believe researchers will be able to its potential while preserving the richness and depth that define qualitative research.

As illustrated in another blog post, qualitative data analysis is often a laborious process, requiring meticulous coding, interpretation, and reflection. Traditional computer-assisted qualitative data analysis software, such as NVivo and MAXQDA, has long been used to help streamline aspects of qualitative data analysis. However, generative AI, and specifically ChatGPT, introduces an additional layer of adaptability, offering real-time feedback and dynamic analytical capabilities. This made us wonder how effective it would be in the qualitative data analysis process.

In our paper, we explore how ChatGPT can function beyond a simple data processing tool by actively participating in the interpretive process. Rather than merely classifying text, we found that ChatGPT could highlight implicit themes, suggest theoretical frameworks, and prompt deeper reflections on the data from both the researcher and participant. However, ChatGPT’s capacity is highly contingent on the researcher’s ability to craft well-designed prompts.

One of the key takeaways from the study is the significance of effective prompt design. We note that ChatGPT’s responses were only as good as the prompts it received. Initially, we found that the ChatGPT’s responses lacked depth or were fixated on single aspects of a topic while neglecting others. By refining our prompts, explicitly defining key concepts, and structuring questions carefully, we were able to guide ChatGPT toward more nuanced and insightful analyses.

We developed a series of 31 prompts to explore our dataset (see the prompts here). This iterative prompting process not only improved ChatGPT’s analytical output but also helped the researcher clarify her own theoretical perspectives. Our study consequently frames this prompt design process as a reflexive exercise, demonstrating how the act of crafting prompts can refine a researcher’s conceptual thinking and analytical approach.

An unexpected yet valuable outcome of using ChatGPT in the research process was its ability to stimulate the researcher’s higher-order thinking. By engaging with the ChatGPT-generated interpretations, the researcher was prompted to critically assess underlying assumptions, refine theoretical lenses, and explore alternative perspectives she might not have initially considered. This process encouraged deeper engagement, pushing the researcher to interrogate her own biases and methodological choices. As a result, the interaction with ChatGPT became an intellectual exercise in itself, allowing the researcher to refine and expand her analytical thinking in ways that traditional methods may not have facilitated as effectively.

One of the most striking findings from our study was ChatGPT’s ability to uncover implicit meanings within qualitative data. For example, when asked about concepts like “illusio” (investment in the socially constructed values within a field), ChatGPT was able to infer instances of this concept even when it was not explicitly mentioned in the data. However, we also found that the ChatGPT-generated interpretations sometimes diverged from participants’ own perspectives. This emphasises the critical role of human oversight. Generative AI lacks self-awareness (at least at the moment!), meaning that its responses must be carefully evaluated. Generative AI can be a powerful tool for organising and prompting analysis, but it is the researcher’s interpretive lens that ultimately determines the depth and rigour of qualitative inquiry.

One of the most innovative aspects of our study is its participatory approach, in which both the researcher and the participant engaged with ChatGPT’s analyses. Instead of using generative AI as a behind-the-scenes tool, the study involved participants in critically appraising the ChatGPT’s findings, thereby decentralising the researcher’s authority over data interpretation. This triadic model (researcher, participant, and ChatGPT) fostered greater participant agency in the research process. By giving participants the opportunity to review and respond to ChatGPT-generated interpretations, we ensured that the generative AI-assisted analyses did not overwrite or misrepresent participants’ lived experiences. This approach not only enhanced the ethical integrity of the generative AI-assisted research but also enriched the depth and authenticity of the findings.

Questions to ponder

What are the potential benefits and risks of using AI tools like ChatGPT in qualitative research?

How can researchers ensure that ChatGPT-assisted analyses remain ethically sound and participant-driven?

Join my 15 subscribers!