I just read another great post from Susanne Friese. I love the way she is forging different ways to combine reflexive qualitative research and GenAI. GenAI surely opens up many new possibilities for qualitative research, including using it for coding and without coding.

But I don't really buy the main argument. Here's why.

My worry with conversational approaches can be centred on this sentence:

"This shift becomes tangible when researchers conduct a theme analysis and see that major categories—previously developed through weeks of coding—reappear as coherent, evidence-linked themes within minutes." 

Braun & Clarke rightfully criticise human researchers for saying "the themes emerged from the text". But this is just the same, isn't it?

"What are the major themes in this document" is often presented as easy, low-level tasks suitable for an AI. But they are not trivial. They are fundamentally creative and meaning-making and involve a whole mass of evaluative judgements. What on earth is a theme? For whom?

Susanne continues:

"Researchers are repositioned. Their central task becomes judgment: asking meaningful questions, evaluating plausibility, integrating context, articulating theory, and remaining reflexively aware of technological mediation."

But you already handed a mass of judgement-making to your AI when you asked it to identify themes. And in particular Susanne does this right at the start of the workflow, in step 1 (Friese 2025).

Coding is presented as a necessary evil, some kind of crutch on the way to sensemaking which we can now throw away. But where does coding come from? Long before GenAI, a researcher could immerse themselves in a set of texts and eventually emerge with any number of declarative, findings illustrated by quotes. Were they too distracted by the first shiny thing they saw in the text? Who knows? Did they pay too much attention to some parts of it and ignore other uncomfortable parts? In order for qualitative analysis to count as any kind of science, different frameworks and guidelines were constructed to ensure that at least some parts of the analysis were more systematic, retraceable (nachvollziehbar) and transparent (and perhaps even reproducible). There are endless ways to do that, to show that the research process was not distracted by the first shiny thing it saw and ignored the uncomfortable parts. A human, who is essentially distractible by shiny things, can delegate parts of the process to "mere coding" in order to partially mitigate that risk. But we cannot mitigate humans' propensity to be distracted by shiny things by delegating tasks to a GenAI which is at least as distractible by shiny things. I know that conversational approaches do bring in other methods and procedures to mitigate these risks, having read Friese (2025), but I think this could be better spelled out in Susanne's post.

I agree with Susanne that GenAI may be bringing about a paradigm shift in qualitative research, also in its relationship to quantitative research. But that paradigm shift can take many forms. The way we use GenAI to scale causal mapping is another really different way. Causal mapping happens to be based much more on coding. We would argue that it is therefore more systematic than one which starts by asking "what are the themes here?"

But there are also surely hundreds of other ways to use GenAI in qualitative research, most yet to be discovered.