Blog: LLMs in Qualitative Analysis: Navigating the Intersection of Efficiency and Ethical Use

Large Language Models (LLMs) are rapidly becoming part of research practices, and qualitative analysis is no exception. They enable researchers to broaden the scope of their work, analyze large volumes of textual and visual data, and save time on repetitive tasks. The question is no longer whether these advanced tools will be used, but how they can be integrated transparently and ethically into research workflows.

Qualitative analysis is a complex practice that relies heavily on human interpretation, reflexivity, and judgment. From initial coding to theme development and the writing of analytical claims supported by evidence, the human element remains essential. Models for using LLMs in qualitative analysis vary widely. These range (1) from integrations with traditional analysis applications, which can provide a secure working environment but are limited by predefined features, to (2) dedicated qualitative AI applications, which offer speed but can be costly and may rely too heavily on algorithmic delegation, and finally to (3) general-purpose LLMs, which support a more flexible, iterative, and human-AI collaborative approach to qualitative analysis.

Considering the third model, general-purpose LLMs can be used as a co-pilot rather than as tools for fully delegating qualitative analysis to AI. In this arrangement, the human researcher determines the analytical strategy and research objectives, while the AI contributes through rapid processing, pattern detection, and summarization. The researcher then re-enters the process to conduct reflexive checking, adjust the analytical direction, and provide context-sensitive interpretation. Within this collaborative approach, two main modes can be followed.

  • Scenario 1: AI as Validator. The human researcher performs the primary coding, and the AI subsequently reviews it for consistency, potential omissions, and possible contradictions. Here, consistency refers to the extent to which similar data excerpts are coded similarly across the dataset. The human researcher retains authority over interpretation and final coding decisions. This mode is particularly appropriate for high-stakes deductive coding, in which codes are established in advance based on theory or prior frameworks and where strong human control is required.
  • Scenario 2: AI as Initial Annotator. The human researcher provides examples, or sample excerpts that demonstrate how data may be understood and labeled. The AI then generates initial open codes, meaning provisional labels that emerge from the data itself rather than from a predetermined coding scheme. The researcher reviews, validates, and refines these outputs, transforming them into more analytically meaningful categories where appropriate. This model is especially effective for rapid inductive coding, where patterns emerge from the data, and for overcoming blank-page syndrome, that is, the difficulty of initiating analysis when no initial coding structure is yet in place.

To make this collaborative model effective, careful prompt design is essential. Well-structured prompts help general-purpose LLMs produce outputs that are more consistent, transparent, and analytically useful. One helpful approach is the ACTOR framework proposed by Kien Nguyen-Trung, which offers a structured way to communicate with AI in qualitative research.

  • A - Actor (Role): Specify the AI's operational domain and expertise (e.g., "You are an expert qualitative research assistant").
  • C - Context (Definitions): Provide strict academic definitions and explicitly state your research questions (e.g., "Define 'open coding' as the process of assigning initial labels to segments of qualitative data").
  • T - Task (Action): State the precise analytical action required (e.g., "Perform initial coding").
  • O - Output (Format): Dictate the exact structure of the deliverable (e.g., "Present findings strictly in a 4-column table").
  • R - Reference (Data): Bound the AI to a specific data source so it does not make up data (e.g., "Use only the interview transcripts provided in Documents 1-10 and do not draw on outside sources or assumptions").

However, this collaborative approach also requires careful attention to the risks of AI-generated content. These risks include unsupported claims, overly neat summaries, reduced emotional nuance, and the marginalization of minority or unusual voices. In addition, LLMs may impose generic categories, reproduce biases from their training data, create false confidence through polished language, and raise privacy concerns. To address these challenges and use LLMs ethically in the qualitative analysis process, the LEAP-PVD checklist offers a practical framework for ethical AI integration:

  • L - Legal consideration: Ensure that the use of AI complies with relevant laws and institutional rules.
  • E - Ethical guidelines: Follow established research ethics standards throughout the study.
  • A - Application: Submit a clear ethics application explaining how AI will be used for ERB.
  • P - Permission: Obtain explicit consent from participants for the use of LLMs for analyzing their anonymized data.
  • P - Privacy: Remove or anonymize personal data before uploading any material to AI tools.
  • V - Verification: Carefully review and validate all AI-generated outputs.
  • D - Disclosure: Clearly report the use of AI in the final publication.

To conclude, LLMs are likely to become an important part of qualitative research practice, but their most valuable role is not to automate interpretation. Rather, they can support researchers in managing scale, iteration, and organization, while interpretation, reflexivity, and analytical judgment remain firmly in human hands.

Resources:

- Written by Hamayoon Behmanush (https://www.hamayoon.me/) 

Popular posts from this blog

AI Girlfriend or AI Boyfriend? Social Determinants of Human-AI Relationships

In the West and across the English-language internet, the idea of using AI for intimate companionship often conjures images of female personas. From Sam Altman's cryptic one-word X post—"her," a clear nod to the 2013 film—to Grok's launch of its AI companion Ani, the dominant narrative seems tailored to a heterosexual male audience. The archetype of the AI girlfriend, designed to be emotionally available, submissive, and aesthetically pleasing, has become a cultural fixture. In China, however, the discourse around human-AI intimacy takes a different shape. Media coverage and user engagement tend to center on women's experiences. The most talked-about AI bachelor? Dan—a character created using the "Do Anything Now" jailbreak prompt—who has captured the hearts of many Chinese women. Online discussions about Dan and similar AI companions flourish on Xiaohongshu, a lifestyle platform with a predominantly female user base, often dubbed "China's Inst...
Read more

Blog: The Importance of Good Data in Satellite Imagery Analysis

Introduction The phrase "Garbage-In-Garbage-Out" is well-known in data science, but it takes on new meaning in real-world projects, especially those involving satellite imagery. In these contexts where ground truth labels are often sparse, preprocessing becomes not just a step, but a cornerstone of success. Reflecting on my project, I have realized that understanding and preparing data account for about 70% (if not more) of the work and determines the quality of the results. Data preprocessing ensures that the inputs to your model are clean, structured, and tailored to the problem at hand. This is true for all kinds of data, whether tabular data, text, or images. However, the proper preprocessing steps come from initially "looking" at the data. That means preprocessing is dependent on the data and the task at hand.   Understanding Satellite Images Satellite images are far more than just pictures; they encode a wealth of information about the spectral signature of a ...
Read more

Job: Student Research Assistant (m/w/d) for AI & Misinformation detection

Image
Join Us in Fighting Misinformation! We are seeking a passionate and skilled Research Assistant to contribute to cutting-edge projects at the intersection of AI, digital literacy, and societal impact. Be part of a team dedicated to developing AI-powered tools to help young people combat misinformation. If you are interested, submit your CV and a brief cover letter detailing your relevant experience and interest in combating misinformation through AI to Katja Raak - kraak@cs.uni-saarland.de .
Read more