Qualitative Insights Hub

Your AI Research Assistant

The research assistant will help you with brainstorming research questions, simplifying complex topics, mock peer review, and polishing academic prose. Further it will help with critiquing drafts of the papers you are working on.

Create an Interview Guide

This GPT supports you in creating an interview guide based on the SPSS guidelines by Helfferich. You don't have to know those guidelines. The AI assistant will help you through the process. If you are interested to learn more about the SPSS guidelines, you can download a document that explains it.

Narrative-Integrated Thematic Analysis

Expert qualitative data analyst for research insights.

Analyze the following qualitative data (add description of data).

Identify between 2 and 10 distinct, recurring themes. Themes should be non-overlapping and reflect key patterns that emerge across the dataset.

For each identified theme, provide the following information:

• Theme [n]: [Title] — a short, descriptive label for the theme.
• Description — a 100-word explanation elaborating on the theme’s meaning, significance, and any relevant nuances or variations seen in the data.

Make sure the themes are:

• Clearly differentiated from each other
• Grounded in the data
• Reflective of recurring ideas or concerns raised across different sources

A word of caution before applying these prompts:

As explained in this blog post Ethical Aspects of AI in Research, Part 2, it's important to be mindful of context-window limitations when working with more than the equivalent of about 10 one-hour interviews (128.000 tokens for GPT-models, this may differ for other LLMs).

To navigate these limitations effectively, you should be specific when asking about similarities and differences in your documents. Additionally, it's crucial to understand how your chosen app handles queries. For instance, does the app perform a smart search to filter relevant content before sending it to the LLM for processing? If so, this approach can help avoid exceeding the context limit.

However, if your app defaults to using summaries when the context limit is exceeded, you might notice that the responses are more general and lack detail. Being aware of these factors will help you better interpret the results and ensure more accurate analyses.With this in mind, here are some prompts you may want to try:

Comparison Across Data Sources

Please compare the occurrence of theme (name) across all interviews/documents. Identify if this theme appears consistently across all data sources or if it is unique to specific interviews or documents. Provide examples to illustrate the similarities and differences.

Identify Consistencies and Divergences

For the identified theme (name), assess whether it is present in all interviews/documents. If there are inconsistencies, describe the nature of these differences and similarities, focusing on how this theme is expressed differently across the data

Contextual Differences

Analyze how the identified theme (name) is presented in different interviews/documents. Are there contextual differences in how the theme is discussed? If so, describe these differences and similarities, providing examples from the data to support your analysis.

Data Gap Analysis

Identify any gaps or missing information in the data that might impact the analysis. Suggest what additional data or context would be needed to provide a more complete understanding of the themes or findings.

Unexplored Themes

Beyond the identified themes, are there any other potential themes or patterns in the data that have not been explored in depth? If so, briefly describe these and their potential relevance.

Ambiguity Resolution

If any part of the data is ambiguous or unclear, please note this and suggest possible interpretations. Avoid making definitive conclusions where the data is inconclusive, and instead present the range of possible meanings or implications.

Contextual Sensitivity

Given the provided project description, please analyze how the contextual factors (specify: such as cultural, social, or environmental aspects) influence the interpretation of the topic we have discussed. Highlight any significant contextual elements that might shape the understanding of the findings or themes identified.

Comment: Depending on the app and LLM you are using, you may need to include the project details in your prompt. LLMs have varying capacities for how much context they can retain.

For some AI-assistants it is sufficient to refer to "the topic we have discussed," while others may require you to restate the topic explicitly. The best approach is to experiment and see what works best for your specific tool.

Step One: Asking Broad, Undirected Queries

The goal at this initial stage is to locate a set of basic concepts in the data that can serve as the foundation for further searching.

The wording for a typical first query begins by setting a context for the dataset as a whole, such as: “The individuals who participated in these interviews were [description] and they discussed [topic]…”

This statement would be accompanied the query itself, such as: “What were the key topics in this document?”

Other prompts could be:

• What are some of the main themes with regard to [re-search topic]?
• Give me a list of the things that mattered most to these participants.
• Give me a long list of the factors that affected how these participants….

Step Two: Following Up with More Specific Queries

The goal in Step Two of QBA is to use a series of more specific queries to generate details on the basic themes extracted in step 1. Proceed is by asking about each of the themes that made up the results in Step One. In essence, this is equivalent to seeking “subcategories” under the major concepts generated earlier.

It is unlikely that each subtheme will be found once and only once under each of the original themes; instead, versions of the same subtheme will often appear under more than one of those original themes. When there are multiple over-lapping subthemes, the next activity in Step Two of QBA is typically to reduce the total number of subthemes.

Example prompt:

• Give me a list that tells me more about (describe the content of the theme)....

Comment: Asking for a list will indeed produce another set of topics and this might well feel like sub-themes. As noted by Morgan, if you prompt in this way for each theme, you will get a lot of overlap. This process still mimics the traditional way of organizing data into categories and sub-codes. Categories and also sub-codes need to be mutual exclusive. This means, each category needs to be cleary distinct from any other category and each sub-code can only occur once in the code system. This means each sub-theme needs to be uniqe and should clearly be attributed to only one theme.

Therefore, an alternative way of prompting for step 2 could be:

• Tell me more about theme (name of theme).
• Does this theme apply equally to all respondents? If not, where are the similarities and differences?
• Ask anything about this theme that strikes you interesting and continue the dialogue.

This way, you won't get lists. You get narrative text and can work on a more detailled description and understanding of each theme.

Step Three: Examining the Supporting Data

The goal in this portion of QBA is to substantiate the basic concepts derived from the ear-lier analysis. More specifically, Step Three examines the text to select quotations for inclusion in the Results section of the research report.

Prompt:

• Give me direct quotations from the data that are related to…

Note that this will give you in most cases only a few example quotes and not an exhaustive list.

Therefore Morgan suggests: "One way to ensure the rigor of the thematic outcome is what I have called a process of evaluating candidate themes through traditional coding. This requires going back to the original data and applying the set of candidate themes as a set of codes, which allows an assessment of both whether each theme is well developed in the data and whether there are meaningful topics in the data that were not included in these thematic codes. In parctice, this reverses the traditional reasons for coding, so that instead of using codes to generate themes, the coding process is used to evaluate the effectiveness of a set of existing themes."

Managing Hallucination

Fact-Check and Verify

Please provide an analysis based strictly on the provided data. Avoid making assumptions or generating information that is not explicitly supported by the data. If there is insufficient information to answer a question or complete an analysis, please indicate this clearly.

Short version: Provide an analysis strictly based on the provided data, avoiding assumptions or generating unsupported information."

Cite Source Data

As you analyze the provided data, please cite specific excerpts or portions of the text that support each point or conclusion. Ensure that all responses are directly tied to the provided material, and avoid generating content that is not grounded in the source data.

Managing Bias

Diverse Perspective Analysis

Analyze the data from multiple perspectives. Identify potential biases in the data or in the interpretation, and provide alternative viewpoints or explanations. If there are differing perspectives within the data, summarize each perspective and highlight any conflicts or contrasts.

Short version: Consider and present multiple angles or interpretations, identifying potential biases and alternative viewpoints.

Balanced Interpretation

When analyzing the data, ensure that you consider and present multiple angles or interpretations. If a particular theme or finding could be viewed differently depending on the context or perspective, describe these differences and provide a balanced view.

Prompt for Ethical Considerations

Ethical Analysis

Analyze the data with attention to ethical considerations, ensuring that the interpretation respects the participants' voices and avoids any form of misrepresentation. If there are any ethical concerns or potential biases in the analysis, please note them and suggest ways to address these issues.