6.2. AI Prompt Library
-
Prompts play an important role in any type of qualitative data analysis working with large language models. Prompts tell the AI how to analyze your data and respond to your queries. They are written in natural language and are easy for non-programmers to understand.
-
In QualCoder, the prompts underlying the AI features are accessible in the user interface. They can be customized to fit your methodological approach and research questions.
-
We have already included some interesting prompts from other qualitative researchers that have been published in various places. We hope that the growing use of AI in qualitative research will be accompanied by a culture of developing, sharing, and discussing sophisticated prompts in the scientific community, so that we can all learn from each other and improve the emerging new methods in the field.
In a very basic sense, a "prompt" is like the question you type into the web interface of ChatGPT. But even in this case it's a bit more complex. In the background, ChatGPT will accompany your question with an invisible "system prompt" that instructs the AI to be a helpful and supportive assistant, not to give medical or legal advice, etc.
Similarly, the prompts in QualCoder are modular and consist of several different elements, some of which you can change and some of which you cannot.
This is the basic structure:
This structure consists of
- fixed elements (shown here in white on blue) that are defined in the source code of the application,
- contextual information (here in green) that you enter at various places in QualCoder's user interface, such as code names and memos, or the project memo with important information about your research,
- empirical data (here in yellow), selected in various ways, and
- the analytic prompt (shown in orange), which you can modify and adapt to your needs.
Analytic prompts are the core of QualCoder's AI features. They instruct the AI on how to analyze your data. These instructions can be quite simple, but can also grow into complex guidelines with detailed descriptions of each step the AI should perform during the analysis, including methodological background, etc. (see Lieder & Schäffer, 2024).1
QualCoder comes with a set of predefined analytic prompts. But you can also define your own - either by clicking on the "Edit" button next to the prompt selection or by navigating to the "Prompt library" located in the "AI" menu. This will open the following window:
The prompts are divided into several categories:
-
Type
- Search prompts are used in the AI assisted coding,
- Code, Topic and Text Analysis prompts are used in the AI Chat window.
-
Scope
- System prompts are defined on the application level and cannot be altered directly. If you want to change a system prompt, you must duplicate it first.
- User prompts are defined by the user and stored on their computer. They are available to any project opened on that computer.
- Project specific prompts are stored with the project files. If you or someone else opens the same project on a different computer, these prompts will be available there as well.
-
Prompt name and description
These are displayed in several places in the user interface where you can select an analytic prompt. The names must be unique within each scope.
-
Prompt text
These are the actual instructions that will be sent to the AI (the orange box in the image above). If you want to share your prompt with someone else, this is the most important information you will need to send over.
All predefined prompts in QualCoder are written in English. However, they can be applied to data in other languages as well.
It is also no problem to write analytic prompts in languages other than English and use them with QualCoder. Most large language models are quite flexible when it comes to mixing different languages.
1Lieder, F. R. & Schäffer, B. (2024). Reconstructive Social Research Prompting (RSRP). Distributed Interpretation between AI and Researchers in Qualitative Research. https://doi.org/10.31235/osf.io/d6e9m
