Objectivity: Researchers are encouraged to conduct their work free of personal prejudices and to adopt a neutral and impartial attitude at all times. It is well known that subjectivity is a strength of qualitative research. What is important with regard to this criterion is the recognition and reflection of possible biases. Careful and accurate documentation: It is essential that all research data is accurately recorded and conscientiously documented to ensure traceability and reproducibility of the research. In order to be accountable for this criterion, the researcher must know how research results were obtained.

Using an interactive approach allows you to collaborate with an AI assistant, directing the analysis rather than letting the AI take the lead. Instead of relying on simple buttons like “Summarize data,” “Extract themes,” or “Code the data,” you can enter prompts and initiate deeper exploration. This interaction enables you to reflect on the results, questioning and probing where necessary. If you suspect that an answer is biased or fabricated, you can highlight this concern to your AI assistant. You can request the assistant to reflect on its response and provide reasoning behind it. Additionally, you can challenge the AI to offer alternative perspectives. If you suspect that the AI has generated incorrect or hallucinated information, you can ask it to re-examine the data and provide supporting evidence.

Below, I'll provide an example of how an interactive analysis with an AI assistant can look. However, to understand why I recommend starting with either a single case or a small subset of your data, you need to be aware of an important current technical limitation with LLMs.A Current Technical Limitation of LLMsAlthough there are many promises about what AI can do for us, there are also current limitations that are often not discussed. One such limitation is the so-called context window. This has been brilliantly explained by Jeff Lagana:“Picture yourself engaged in a conversation with a friend. Initially, you discuss plans to have dinner together and agree to meet at 6pm. You then delve into your respective days, their family matters, and your weekend plans. As you're about to part ways, you ask again about the dinner plans. However, to your dismay, your friend has no recollection of ever discussing dinner or making any arrangements to meet. It can be frustrating to realize that someone has seemingly forgotten something that was previously discussed.While this analogy is merely aggravating in real life, it highlights a fundamental limitation of today's large language models. This limitation becomes particularly significant when dealing with generative AI and processing extensive datasets.

Repeat steps 2 and 3 for each research question you want to investigate with your AI assistant. If your data set is larger than what fits into the context window, continue exploring your first research question with the next set of data. You could start by asking the same initial question and build from there, or your entry point could be some abstract concepts you developed when working with the first set of data.To give you an example, if we were to analyse the next set of five interviews in the work-life balance study, I could use the two types I have developed as entry points and ask the AI assistant to identify these types in the other interviews as well. I could then investigate the role of flexibility and explore further with the AI assistant whether other types beyond those identified in the first set of interviews are present in the data.

Researcher: I see Kazumi and Tatjana similarly; they both seek more separation to better manage their jobs, childcare, and self-care. There is some overlap with Amadi, particularly regarding reduced stress following a breakup. David and Arne share a different approach, as they love their work so much that they don't want strict separation since their job is their life. However, Arne lacks the flexibility that David has. Flexibility is a common theme among all five respondents, with each believing that more flexibility would lead to better balance. Based on this, create three types and highlight the overlaps.