"Advanced Data Analysis" with ChatGPT-4 | Cross-case and Within-case thematic analysis

The core takeaway is a practical workflow for using ChatGPT-4 (via ChatGPT Plus’s “Advanced Data analysis” feature) to perform both within-case and cross-case thematic analysis—first extracting themes from each interview separately, then merging them into a shared thematic framework and finally probing differences by gender and age. The value is speed and structure: individual transcripts get summarized into leadership-style, challenge, and strategy themes, and those outputs then feed into a cross-case synthesis that can highlight patterns across participants.

The process starts with preparing interview transcripts that already include demographic markers. In the example, four leader interviews are uploaded, with each transcript labeled for gender (female/male) and age (e.g., “female 49 years old”). The demographic tagging matters because it gives the model a basis for later comparisons, not just a way to summarize leadership narratives.

A targeted prompt then instructs the system to analyze each transcript one at a time and list what can be said on key topics—leadership style, challenges, strategies, and related guidance. The “Advanced Data analysis” option processes the file directly (avoiding copy-paste) and returns a breakdown per participant, which functions as the within-case analysis. Notably, the workflow also produces extra content such as advice for aspiring leaders even when that wasn’t requested, suggesting the model may expand into adjacent themes when it sees leadership-related material.

After the within-case outputs are generated, the workflow shifts to cross-case analysis. The next prompt asks for a “common thematic framework” that combines the separate participant frameworks into one integrated set of themes. The resulting synthesis includes shared categories such as leadership styles, challenges, and organizational challenges, and it may also comment on differences that emerge across gender and age.

To sharpen the comparison, a follow-up prompt explicitly requests insights related to gender and/or age. This step yields theme-level comparisons that attribute what females and males emphasized, and it can incorporate combined gender-and-age insights. The presenter emphasizes that results can be inconsistent: running the same prompt on the same data may produce different levels of specificity, so repeating the request can sometimes yield a better comparison.

A key methodological caution runs through the workflow: overly strict instructions that demand full “thematic analysis” rigor (including detailed coding stages) can overwhelm the model or cause it to fail. The workaround is to use simpler, “common sense” language—asking for analysis and topic-based listing rather than demanding the entire formal coding pipeline in one go. The tool is positioned as a strong foundation for further investigation, not a replacement for researcher judgment, since it can be random and should not be relied on to complete an entire thesis end-to-end.

Overall, the transcript presents a repeatable, prompt-driven approach: within-case extraction per participant, cross-case synthesis into a shared framework, and targeted demographic comparison—while managing model limitations through pragmatic prompting and iterative refinement.