I am using ChatGPT to analyze my qualitative data. My goal is to find the key themes in the data without the time-consuming and tedious process of coding.

As for hallucinations, I have not encountered this problem, but there is no substitute for being familiar with your data and your field of research.

Thank you. Can you share queries you use and which platform you found most useful. Do you include this in your methodology description. Also, I agree with the expertise aspect.

I use ChatGPT 3.5 (I have compared it with ChatGPT 4 and found little difference for my purposes). In particular, I use ChatDOC as an interface because it marks segments in the original data that match up with its responses. That is helpful for finding quotations to use in my Results section.

I start with a very general prompt such as, "Give me a list of the major themes in this data." The goal is to do a "top-down" analysis (rather than "bottom-up" such as coding) that does not rely on substantive input from myself.

FYI I have actually chatted with ChatGPT 4 about alternatives to coding in qualitative data analysis, and it spontaneously came up with the label Query-Based Analysis for what I do, and the program even started referring to as QBA. But when I asked it for any references or citations it did not have any.

Hi David. I have no experience of chatgpt besides the silly question now and then. Could you please tell me how do you feed it your qual data sets? Do you copy and paste your transcripts one by one?

Hi David, Have you had any issue from journals when you report that you used ChatGPT for the thematic analysis of the text? Do you explicitly state in consent forms that verbatim text may be uploaded to ChatGPT? Are there any rules about that in place yet? Haven't done this myself yet, but curious how others are approaching issues of privacy and how journals are responding?

I have been working on methodological articles about using ChatGPT in qualitative data analysis, so I have not submitted any straight empirical articles for publication. I have attached a late-stage draft of an article that is forthcoming in the International Journal of Qualitative Methods.

My approach (which I am developing in a follow-up article) does not use Thematic Analysis -- or coding in general. Instead, I use a query-based approach that asks the software to suggest themes, which I then query further. This requires strong familiarity with the data to judge the quality of the AI's responses, but I have found it to be right on target for the data I am giving it.

All the analyses I have done are with older data sets, so I haven't run into the issue of informed consent. Depending on which program you use, you may have the option not to allow inclusion of your data in their database.