Watch this Before you use AI in your dissertation research / Qualitative data analysis with AI

AI-assisted data analysis tools are increasingly popular for qualitative dissertation work, but they often fail at one non-negotiable requirement: producing a transparent audit trail of how results were reached. That gap matters because qualitative validity depends on showing that themes emerged from a structured, traceable process—especially coding—rather than from intuition or “it feels like” interpretations that can reflect bias.

In academia, the expectations for methodological rigor are particularly high. Students and researchers aiming to publish in academic journals are typically required to demonstrate credibility and trustworthiness through transparency. In practice, that means being able to document each step of analysis so an external reader—supervisor, examiner, or reviewer—can follow how codes were created and how those codes were developed into themes. Without that trace, it becomes harder to argue that findings are credible, and easier for concerns about researcher bias to surface, even when intentions are good.

A detailed coding approach is presented as the backbone of both rigor and validity. Coding provides structure and consistency: if the same person (or another analyst) repeats the process, the resulting codes and themes should be broadly similar because the analysis is built step-by-step rather than improvised. This repeatability is framed as a practical safeguard against bias and a way to strengthen confidence in the final thematic conclusions.

Online AI analysis platforms, by contrast, are described as fundamentally mismatched to audit-trail expectations. Even when they generate accurate or impressive outputs, they generally do not expose the intermediate reasoning or the coding workflow. Users may upload data and receive results, but they cannot inspect, reproduce, or expand the analysis in a way that supports transparency. The result is a “black box” problem: useful outputs without the documentation needed for academic scrutiny or for independent verification.

ChatGPT is singled out as an example where some audit-trail-like control can be created through an interactive workflow—issuing commands and monitoring the back-and-forth. Still, consistency is flagged as a limitation: the same prompts may not reliably yield the same results, which undermines the stability that validity demands.

The practical recommendation is to prefer professional qualitative data analysis software that keeps the analyst in control of decisions and preserves a full audit trail. Established tools named include MAXQDA, NVivo, and ATLAS.ti. Which one to choose depends on access: researchers affiliated with an institution should use the platform their institution provides. For independent researchers without institutional access, the guidance favors ATLAS.ti as a currently reliable, professional, and user-friendly option, while acknowledging that the core decision should be driven by the ability to maintain transparency and rigor throughout coding and theme development.