A letter signed by 419 researchers rejects the use of generative artificial intelligence for reflexive thematic research. I discuss issues of phenomenology, epistemology, and ethics.

In December 2025, Tanisha Jowsey (social scientist in medical education), Virginia Braun (psychologist), Victoria Clarke (psychologist), Deborah Lupton (sociologist), and Michelle Fine (psychologist) published a letter signed by 419 qualitative researchers from 32 countries. These experts reject the use of generative artificial intelligence (GenAI) for reflexive thematic research. This includes interviews, focus groups, and other phenomenological methods. In this post, I cover the key points of the letter, as well as issues of phenomenology (subjective understandings of the world) and epistemology (how we understand facts), which GenAI cannot address.

The researchers give three primary reasons for rejecting GenAI in qualitative research:

1. GenAI lacks understanding of human meaning, which is the central focus of qualitative methods
2. Interpretations of social data require humanity
3. GenAI has been developed through exploitative measures.

I support the sentiments of this letter.

The statement speaks to the importance of phenomenology. This concept describes the study of subjective understandings of the world, and lived experiences of individuals. Qualitative methods are centrally concerned with how people learn about, and engage with, the world around them. GenAI is unable to think critically about subjectivity, because artificial intelligence (AI) programs do not think. By design, these programs amalgamate information, stolen from online sources, and generate “new” regurgitations from these pre-existing sources, rather than weighing up the nuances of human experience through social theory and methods.

@professorcasey

What is an AI reasoning model? Part 13 of how large language models actually work: chain of thought and “reasoning” for LLMs.

♬ original sound – Professor Casey Fiesler

The statement also evokes debates about epistemology. That is, the theories and methods of knowledge, and the ways in which we understand “facts,” “truth,” and “objectivity.” For example, the types of questions we ask, the theories we use, the choices we make when designing a study, what we leave out, and so on. Qualitative researchers address their epistemological standpoint (their background and training) shapes their research. GenAI does not know to do this, or how to do this. It simply spits out summaries, based on the assumptions baked into their programming.

Elsewhere, I have published my own research on the methodological and ethical issues with generative AI.

My research recognises that AI may have useful applications, and that sociologists and other social scientists, can help to shape the development, regulation, ethical guidance, evaluation, and transparent governance of AI.

While some applications of AI may have some merits, I have shown that generative technology is flawed. It has a high rate of error, and contains a plethora of biases, which in turn reproduces various inequalities, such as racism and sexism. I also demonstrate that AI models cannot solve complex problems, and this technology cannot meaningfully replicate the work of social scientists, the humanities, and creative workers.

Research shows that 95% of companies who have invested in AI technology have seen no return on investment. Studies also show that 40% of staff find AI does not save them any time at work, and, similarly, 40% of workers waste over three hours a month correcting AI mistakes. Research shows AI “hallucinations” (made up information, sources, and answers that are presented as facts) are rampant, appearing in up to 80% of legal cases.

Generative AI is aggressively promoted, and difficult to escape, as many companies, workplaces, services, and products have forced it upon staff and customers. While I don’t use AI, including in any of my writing or research, nearly every tool I use daily has incorporated AI, from document software, to graphic design, to social media, to the website I’m writing on. Some of these features are optional, others are a hassle to turn off, or require manual closing every session or for each new update.

While AI seems ubiquitous, it is not inevitable.

How, when, and why we use this technology is a choice in some individual cases, especially in research. (I note that some workers are forced to use AI, such as in customer service.) When it comes to qualitative research, generative AI is nonsensical, lacking the basic component of this methodology (humanity), and unethical.