Sage Journals: Discover world-class research

Abstract

Rapid advances in artificial intelligence (AI) technologies, especially in consumer-available large language models (LLMs), have spurred efforts to automate qualitative data analysis. For researchers new to qualitative research, existing work rarely explains how research values and the different approaches to qualitative research shape AI-assisted data analysis methods. We aim to open a discussion about the role of qualitative research values and approaches in AI-assisted data analysis and how this may shape the application of these methods. We begin by outlining the two approaches to qualitative research in the literature (“Small-q” and “Big-Q”) and how these guide the alignment between research values and the use of methods. We then highlight the relevance of these approaches for developing AI-assisted analysis methods by reviewing seminal work that uses LLMs for qualitative analysis. Building on the two approaches to qualitative research, we propose an approach-based model that can be used to understand the alignment between the values of qualitative research and the use of methods. Additionally, we position AI within this model and propose questions to help understand AI-assisted analysis within these approaches. Next, using exploratory examples, we underscore the importance of these considerations by highlighting how alignment with the different approaches and their values may shape the application of AI in assisting reflexive content analysis. These examples are not intended as definitive guides; rather, they illustrate the influence of aligning with each approach. Thus, to support researchers interested in applying these AI-assisted analysis methods in practice, we provide references to further literature. To conclude, we present a summary of ethical considerations and future directions. While AI-assisted analysis will not be suitable for all qualitative research, this paper initiates a discussion about the theoretical challenges of applying AI technologies in qualitative analysis and demonstrates how these considerations change the use of AI-assisted analysis.

Keywords

qualitative research artificial intelligence AI-assisted analysis paradigm qualitative analysis

Introduction

The theoretical values underpinning qualitative research are fundamental to understanding and analysing qualitative data (Braun & Clarke, 2022b, 2023; Collins & Stockton, 2022; Kidder & Fine, 1987). These values shape nearly every aspect of a study, and during analysis they determine how knowledge is understood and evaluated (Carter & Little, 2007; Chamberlain, 2000, 2015; Collins & Stockton, 2018; Lincoln et al., 2011; Pascale, 2011). The rapid development of artificial intelligence (AI) technologies, particularly consumer-available large language models (LLMs), such as ChatGPT, has led to a growing number of papers reporting the use of AI-assisted qualitative data analysis methods.

These seminal studies show how AI-assisted analysis methods can be applied in qualitative data analysis. However, current research has not discussed in detail the foundational values of the different approaches to qualitative research and how these shape different ways of using AI-assisted analysis methods. The values underpinning qualitative research, also often referred to as paradigms (Lincoln et al., 2011), are fundamental to understanding different perspectives on how data analysis is conducted and evaluated.

Paper Outline

In this paper, we aim to open a discussion on the role of qualitative research values in AI-assisted data analysis and how this may shape the application of these methods. We begin by outlining the two approaches to qualitative research that provide an understanding of the consistent alignment between different research values and the use of methods. Within this context, we discuss the growing interest in AI-assisted analysis and outline seminal work demonstrating the use of LLMs for analysis. We highlight that current work has largely not considered how the underlying values of the two approaches to qualitative research influence the practical implementation of AI-assisted analysis and the evaluation of these methods. To address this gap, we introduce an approach-based model of qualitative research and present six questions to help researchers understand how aligning with the values of the two approaches influences their application of AI-assisted analysis methods. We then present exploratory examples using reflexive content analysis to illustrate how considering these two approaches to qualitative research yields different conceptualizations of AI-assisted analysis. Finally, we discuss the ethical and future directions of AI-assisted analysis.

Background

The Two Approaches to Qualitative Research

Before we can discuss how research values shape the use of AI-assisted analysis methods, it is important to understand the alignment between these values and the processes of doing qualitative research. Despite its widespread adoption and growth, qualitative research is often misconceived as a single, unified approach that merely involves using textual, visual, or audio data (Bush et al., 2020; McCosker & Sendall, 2015; Natow, 2022; Povee & Roberts, 2014). In reality, qualitative research has a rich history, from quantitative-based textual analysis in the 19th century (Harwood & Garry, 2003) to diverse and interpretive approaches that emerged during the late-20th century (Jovanović, 2011).

Rather than viewing qualitative research as a single unified approach, it is instead useful to distinguish between the two prevailing approaches. These two approaches are fundamental to understanding differences in the values of qualitative research and how qualitative analysis is understood, conducted, and evaluated across disciplines. This distinction, first introduced by Kidder and Fine (1987) and further developed by Braun and Clarke (2019, 2022b, 2023), delineates two approaches to qualitative research known as “Small-q” and “Big-Q.”

Small-q approaches to qualitative research utilise positivist values that assume reality is in some way objective and can be observed, measured, and understood. Based on these values, Small-q approaches employ ways of doing qualitative research and analysis methods that are structured and statistical, designed to objectively observe and measure phenomena and draw on metrics such as validity and reliability to assess findings. For example, researchers using a Small-q approach might conduct structured textual surveys to identify patterns within a population and validate their analyses using metrics like validity and reliability.

In contrast, Big-Q approaches draw on non-positivist values that emphasise subjective interpretation and the situatedness of data in analysis (e.g., post-positivism, critical realism, and constructionism). These values highlight the subjective nature of reality, focusing on understanding individual experiences, social contexts, and multiple perspectives. This supports the use of flexible, interpretive, and reflexive methods that leverage human subjectivity in understanding data (Braun & Clarke, 2022a; Collins & Stockton, 2018; Nicmanis, 2024; Paulus & Marone, 2024). Rather than validity or reliability, Big-Q approaches focus on metrics such as trustworthiness to assess analysis quality (Guba & Lincoln, 1982). For example, researchers adopting a Big-Q approach might conduct unstructured interviews to perform deep interpretive analyses, in collaboration with participants, aiming to understand their lived experiences and the complex social contexts in which they are embedded.

When positivist values and methods are incoherently combined with non-positivist methods and values, it has been argued that this results in a “confused q” approach (Braun & Clarke, 2021). For example, consider a researcher evaluating the clinical effectiveness of a drug. Such a process is grounded in positivist values, which hold that objective truths about the reality of the drug can be empirically proven, and thus, it would be inappropriate to employ non-positivist methods such as thematic analysis or unstructured interviews with a small sample of participants. Conversely, a researcher exploring the socially constructed nature of participants’ lived experiences would find a textual analysis that only quantifies the frequency of isolated words, without contextual depth, to be insufficient to understand the issues experienced by participants.

Fundamentally, these two approaches provide an understanding of the coherent ways in which different qualitative research values can be aligned with different methods for doing qualitative research. During qualitative analysis, adopting either positivist or non-positivist research values fundamentally shapes the use of analysis methods and how knowledge is generated, understood, and evaluated (Chamberlain, 2015). Consequently, these approaches allow researchers to recognise how their specific research values shape the way they use and understand analysis methods, and, ultimately, AI-assisted analysis.

Current Work in AI-Assisted Qualitative Analysis

Qualitative research and data analysis are resource- and time-intensive, which students, researchers, and educators often perceive as a substantial limitation of this form of inquiry (Bush et al., 2020; Povee & Roberts, 2014). To mitigate these demands, technological solutions such as automation have long been explored (Marshall & Naff, 2024; Webb, 1999). In areas such as the transcription of recorded audio data (Bokhove & Downey, 2018), and quantitative approaches to content and semantic analysis (Franzosi, 1990), “non-AI” computer automation has been adopted to reduce these demands. However, the recent release of LLMs such as ChatGPT, which feature intuitive chat interfaces capable of generating human-like responses from textual prompts, have sparked growing interest among scholars in the implications of AI for research (Dwivedi et al., 2023).

Newly available LLMs have been promoted as a way to gain deeper insights into qualitative data compared to previous automation methods, due to their enhanced ability to understand and interpret patterns (Chubb, 2023; Hitch, 2024; Marshall & Naff, 2024; Paulus & Marone, 2024; Tschisgale et al., 2023). These LLMs do not understand language as humans do; instead, they interpret statistical patterns to produce natural-looking responses. ChatGPT specifically is a generative pre-trained transformer (GPT) that uses statistical neural networks to produce human-like responses (see Brown et al., 2020 for further information). While other AI-based tools and LLMs, such as IBM Watson (Lee et al., 2020), exist, the ready availability and ease of use of consumer-available LLMs like ChatGPT have gained substantial attention in the qualitative research literature. In the sections that follow, we discuss seminal work demonstrating and evaluating LLMs for AI-assisted qualitative data analysis, highlighting limitations in examining qualitative research values and approaches.

Existing studies demonstrating AI-assisted data analysis have largely explored what would be considered Small-q approaches to AI-assisted analysis, drawing on metrics such as reliability and validity to understand analyses. A key focus of this research has been the use of AI to Assist coding. Tai et al. (2024) demonstrated the recursive use of ChatGPT for the deductive coding of interview data, finding AI coding comparable to human coding and noting that the use of multiple AI analysis iterations may enhance reliability. Similarly, Bano et al., 2023 compared AI and human deductive coding of product reviews using Google’s Bard and ChatGPT, finding modest levels of inter-rater reliability between human and AI-coded data. Beyond deductive coding, Theelen et al. (2024) applied ChatGPT to both structured and unstructured data, achieving results comparable to human inductive coding.

Current work has also explored the application of LLMs to specific analysis methods within a Small-q context, including inductive thematic analysis and content analysis. Paoli (2024) found that inductive thematic analysis conducted by ChatGPT produced comparable themes to those found in earlier studies of open access datasets. Further work by Paoli and Mathis (2024) explored the potential of a mathematical approach to evaluate the validity of inductive thematic analysis performed by ChatGPT. One of the most thorough investigations of AI-assisted content analysis was conducted by Bijker et al. (2024) who evaluated the potential of ChatGPT to assist with the analysis of forum posts about sugar reduction. They compared both inductive and deductive methods of qualitative content analysis. Instead of comparing AI-generated coding schemes with human-developed ones, they conducted a series of 10-by-10 trials which measured inter-rater reliability by aggregating 10 different coding schemes applied to a dataset across 10 ChatGPT conversations. The study found that ChatGPT performed better at producing inductively developed coding schemes than deductively unconstrained or structured coding schemes, with the AI showing promise as a second coder.

Only one paper has explored a Big-Q aligned approach to the use of AI-assisted data analysis. Hitch (2024) explored the utility of ChatGPT in aiding a reflexive thematic analysis at various stages of analysis using sample interviews published in a newspaper article. This work did not focus on positivist concepts such as reliability, bias reduction, accuracy, or objectivity; rather, Hitch proposed working with AI alongside the traditional analysis process to engage reflexively with their work using the AI to help question assumptions and understandings. Hitch highlighted the ability of these AI models to assist in coding and theme development, not as a tool to conduct the analysis, but rather as an augmentation to assist the analytic process.

The Relevancy of the Two Qualitative Approaches for AI-Assisted Analysis Methods

While this prior research offers valuable demonstrations and practical guidance for AI-assisted data analysis methods, it has yet to examine in detail how different qualitative research approaches and values shape the use of AI-assisted analysis methods. Scholars have cautioned against privileging specific methods over these important considerations, stressing that research values are fundamental to understanding how qualitative data analysis methods are understood, used, and evaluated (Carter & Little, 2007; Chamberlain, 2000; Tschisgale et al., 2023). Consequently, current research has left aspects of AI-assisted data analysis methods under-explored, and this may lead to research approaches becoming confused. This is most evident in the development of AI-assisted tools for Big-Q aligned qualitative research and the evaluation of analysis quality.

Big-Q approaches to data analysis draw on non-positivist values and use these to leverage human subjectivity in understanding the contextualities of data (Braun & Clarke, 2022a; Collins & Stockton, 2018; Paulus & Marone, 2024). Such methods are often likened to artistic processes, where researchers creatively interpret and represent phenomena (Chung-Lee & Lapum, 2024; Collins & Stockton, 2022; Nicmanis, 2024). However, software developed for AI-assisted data analysis often presents these tools in ways that undermine the key values of Big-Q approaches by removing the researcher and the importance of their subjectivities from descriptions of the analysis process (Paulus & Marone, 2024). This potentially creates confused research that lacks depth and fails to capture the nuanced and subjective understandings required by these types of research (Braun & Clarke, 2022b; Carter & Little, 2007; Chamberlain, 2000). The current focus on positivist values in AI-assisted data analysis methods has not allowed for exploration of how AI could be used to assist non-positivist Big-Q methods (Paulus & Marone, 2024).

Additionally, both Small-q and Big-Q approaches to qualitative research employ markedly different metrics to evaluate analysis quality. Small-q approaches to data analysis rely on positivist research values and therefore focus on metrics such as validity and reliability to assess analysis quality (O’Connor & Joffe, 2020). On the other hand, these metrics conflict with the non-positivist values of Big-Q approaches, which instead draw on metrics such as trustworthiness to assess analysis quality (Guba & Lincoln, 1982).

Although Small-q and Big-Q approaches necessitate fundamentally different metrics for evaluating analysis quality, most previous studies of AI-assisted data analysis have only considered positivist metrics such as inter-rater reliability (Bano et al., 2023; Bijker et al., 2024; Tai et al., 2024; Theelen et al., 2024) or other mathematically derived measures (Paoli & Mathis, 2024). Consequently, outside the work of Hitch (2024), non-positivist Big-Q applications of AI and quality metrics remain under-explored. This oversight risks reinforcing the misconception that qualitative research merely involves analysing qualitative data from one perspective, rather than reflecting two distinct approaches with unique values and traditions.

Not differentiating between the two approaches to qualitative research does not mean that current AI-assisted data analysis methods lack merit; these methods hold much promise for qualitative research. However, consideration of qualitative research values and approaches enables a more detailed exploration of AI-assisted data analysis that respects the traditions of different methods. To engage with both qualitative research approaches and values, an approach-based model is required that also allows for AI to be clearly situated within each approach.

An Approach-Based Model of Qualitative Research

While the Small-q and Big-Q distinction provides great explanatory power, it does not offer clear guidance on the considerations that underpin the values of each approach or how AI can be integrated with these values. To better understand the position of AI-assisted data analysis, we build on the original distinction drawn between the two approaches to qualitative research (Kidder & Fine, 1987) to present a detailed approach-based model of qualitative research.

As seen in Figure 1, drawing on the original distinction between Small-q and Big-Q approaches, which are determined by the alignment between values and methods for doing qualitative research, the approach-based model consists of two main components: the research values of the project, and the actual methods or practices for doing qualitative research. As discussed above, the alignment of these two components defines qualitative research as either a Small-q or Big-Q approach.

Figure 1.

The Approach-Based Model of Qualitative Research Showing Key Considerations in Research Values and How Qualitative Research is Conducted

The two main components of the approach-based model each consists of subcomponents that represent fundamental considerations for understanding the values of a project (epistemology and ontology, methodology, and a conceptual framework) and how the qualitative research is done (your methods and outcomes). This model does not imply that qualitative research is linear; rather, as the arrows in Figure 1 indicate, how each component is defined in a project can reciprocally influence the others to achieve consistency between values and the use of methods.

When aligned appropriately, the values of a project and the act of doing qualitative research work together to create cohesive and consistent qualitative research aligned with either a Small-q or Big-Q approach. In qualitative analysis, a Small-q aligned approach draws on positivist values, supporting statistical or structured ways of doing qualitative research. Conversely, a Big-Q approach draws on non-positivist values, supporting exploratory, artful, reflexive, and interpretive ways of doing qualitative research. In the following sections, we detail each component of the approach-based model and its associated subcomponents, and demonstrate how the use of AI-assisted data analysis methods can be integrated within this model.

Research Values

Our understanding of differing research values is based on what Collins and Stockton (2018) define as a theoretical framework, which they argue is the fundamental basis for any qualitative research project. While a number of alternative definitions exist to explain the values of qualitative research (Carter & Little, 2007; Chamberlain, 2000), Collins and Stockton’s (2018) definition of what we term research values provides the clearest and most well-developed account of these values.

The research values of a qualitative project consist of three subcomponents: epistemology and ontology, methodology, and a conceptual framework. How each of these subcomponents are considered in a qualitative research project ultimately determines whether research values align with either positivist or non-positivist ideals.

Epistemology and Ontology

Although often overlooked, epistemology and ontology are foundational to research values (Carter & Little, 2007; Chamberlain, 2015). Epistemological and ontological dispositions explain how researchers view the world and how knowledge can be produced (Collins & Stockton, 2018). The ontology of a researcher is their view of what exists, whereas epistemology represents their views on how knowledge is produced and how this knowledge can be justified or validated (Pascale, 2011). These two concepts shape what can be known, how we come to know it, and what knowledge is considered valid, thereby influencing how prior research, methodology, and methods are understood in a project (Carter & Little, 2007; Chamberlain, 2000, 2015).

For example, positivist research draws on an ontology that sees the world as having concrete, knowable truths and an epistemology that dictates that these truths can be discovered through systematic observation and measurement. In contrast, research adopting non-positivist ontologies generally views reality as fallible or constructed, drawing on epistemologies that view knowledge as interpretive, context-dependent, and created through human interactions and experiences.

Methodology

Epistemology and ontology are closely related to the second subcomponent of the research values of a project: its methodology. While the terms method and methodology are often mistakenly interchanged, methodology refers to the logic or rationales used for selecting specific methods and techniques for a research project, whereas methods are the specific procedures and techniques used to collect and analyse data (Carter & Little, 2007; Chamberlain, 2015; Collins & Stockton, 2018; Pascale, 2011). Methodology is best seen as specific plans for how the research processes should be conducted (e.g., qualitative description) and specifies a rationale for the design of a study and the selection of methods. In other words, methodology is to a recipe as methods are to the cooking techniques used to create a dish.

Conceptual Framework

The last subcomponent of the research values of a project is the conceptual framework. In the case of qualitative research values, the term conceptual framework refers to existing knowledge of the topic being studied, prior theories of phenomena, and the motivations for conducting the project (Collins & Stockton, 2018). The inclusion of a conceptual framework extends beyond common paradigm-based understandings of qualitative research (Lincoln et al., 2011) and allows us to examine how prior research shapes the values of a project.

Research is not conducted in isolation; it is shaped by previous scholarship, and it is important to consider and understand how prior research influences the values of a project and whether it aligns with positivist or non-positivist values and a Small-q or Big Q approach. Typically, conceptual frameworks that draw on existing theory and empirical evidence to formulate testable research questions reflect a Small-q orientation. In contrast, Big-Q approaches tend to emphasize exploratory questions derived from both experiences and evidence, aligning more closely with non-positivist perspectives. This means that a conceptual framework is particularly relevant to the application of AI in qualitative research. This is because the context within which the research is conducted, and prior theories and understandings of phenomena, are important in determining the utility and acceptability of AI assistance.

Doing Qualitative Research

Methods

The second component of the approach-based model consists of the techniques or actual processes for doing qualitative research. The first subcomponent, methods, refers to the specific activities undertaken in analysis, data collection, and data management (Carter & Little, 2007; Pascale, 2011). Examples include approaches to sampling, interviews, focus groups, transcription, and the analysis of data. In Big-Q approaches, the use of methods is exploratory, artful, reflexive, and interpretive, valuing subjectivities on the part of the researcher and participants to construct analyses. In contrast, Small-q qualitative research approaches employ more structured, standardised, and statistical methods, often using predefined categories and procedures to collect and analyse data.

Outcomes

The second subcomponent of doing qualitative research consists of the outcomes of a project. Outcomes refer to the final product of the research project and represent the synthesis between the research values of a project and the application of research methods to create cohesive knowledge.

Defining AI Within the Approach-Based Model of Qualitative Research

The use of AI within research and the approach-based model can be understood through a dual-role framework, with AI being viewed as either something that is studied or used to conduct research (Table 1). When studying AI, the main objectives are to understand how the technology functions, advance its development, evaluate its capabilities and limitations, and assess its contextual implications. Conversely, when using AI as a tool or method within a research project, the focus shifts to examining its role in the research process itself: how it contributes to data collection and analysis and the effects it may have on these processes and outcomes.

Table 1.

The Dual-Role Framework of AI With Differences Between AI as a Method or Construct

Element	Studying AI	Using AI
Goal	Investigating and understanding AI itself, including its principles, mechanisms, capabilities, limitations, and implications	Using AI tools as a method to enhance, conduct, or assist the research process
Ontology	Exploring the fundamental nature and existence of AI, defining its essence, categories, and what AI can know	Considering how the use of AI tools influences what can be known and what exists within the research context
Epistemology	Understanding what knowledge AI can produce, how we can acquire knowledge about AI, and how this knowledge is validated	Exploring how AI tools contribute to the generation of knowledge in research, how we acquire knowledge through AI methods, and how this knowledge is validated
Methodology	Examining different AI models, how AI processes information, how AI is trained (e.g., LLMs), and methods for evaluating AI	Considering how different AI models integrate with research methodologies and influence research design
Conceptual framework	How prior knowledge and research understand AI as a construct	How and why the research context would benefit from the use of AI tools
Methods	Employing specific techniques such as algorithm analysis, simulations, experimental setups, and testing to explore or advance AI functionality	The specific AI tools and techniques employed to assist in data collection, analysis, or other aspects of the research process
Outcomes	Advancing theoretical and technical knowledge of AI, contributing to the development of new AI technologies, and informing applications, ethics, and policies related to AI	Improving research findings using AI tools or gaining new insights through AI-assisted methods

Note. AI = “Artificial Intelligence,” LLM = “Large Language Models”.

When using AI to assist qualitative analysis, it can be viewed within the approach-based model of qualitative research. As illustrated in Figure 2, when AI-assisted data analysis is considered as a method in a qualitative research project, it involves epistemological and ontological assumptions about what can be known through the application of these methods, how this knowledge is acquired, and how it is validated. This aligns with a conceptual framework that outlines the reasons for employing AI and its suitability for addressing specific research questions. Additionally, we must consider how the methodologies used in developing and applying AI technologies (e.g., LLMs) integrate with the methodologies we use in qualitative research to guide the use of methods (e.g., qualitative description or interpretative methodologies). Depending on how the research values and the use of AI as a method are oriented in a project, AI-assisted data analysis can be aligned with either a Small-q or Big-Q approach.

Figure 2.

The use of AI as a Method Superimposed on the Approach Based Model. Note. AI = “Artificial Intelligence”

Questions to Integrate AI-Assisted Data Analysis

Although Figure 2 outlines how AI-assisted data analysis can be viewed as a method, when using AI researchers must critically engage with their research to ensure that AI-assisted methods reflect their project’s underlying values and remains consistent with either their Small-q or Big-Q approach. To aid this process and to explore how the choice of research approach and values influences AI-assisted analysis methods, we present six questions. These questions are based on the approach-based model of qualitative research and should be seen as a starting point to help researchers understand how their use of AI-assisted analysis methods aligns with their research values:

1. Is the use of AI-assisted analysis justified by your conceptual framework ? The use of AI should align with your study’s aims, benefit existing research, and expand analytic possibilities for your work in ways consistent with the traditions of a Small-q or Big-Q approach. In Small-q research, research questions are testable, hypothesis driven, or highly quantitative in nature (e.g. quantifying the types of language used) and you must justify how your use of AI can support this based on your research area. Conversely, in Big-Q research, research questions are exploratory, and you must justify why AI can help explore your data or contribute to your reflexive understanding of the data.

2. How does the ontology and epistemology of your research influence the interpretation and use of AI ? For example, in Small-q research grounded in positivist aligned values, AI can be viewed as a tool to generate objective, replicable codes that quantifiable observable patterns. Conversely, In Big-Q research informed by non-positivist aligned values (where reality is viewed as subjective, constructed or fallible) AI can be viewed as a situated lens that helps craft richer, reflexive insights. For example, in the case of Big-Q research, if you are conducting research within a social constructionist paradigm, you may consider how the social biases inherent in AI training are understood in your research project.

3. Does the methodology of your qualitative research align with your use of AI? For example, the statistical prediction that is the basis for modern LLMs aligns well with positivist-oriented methodologies that specify structured and statistical methods of Small-q analysis. For Big-Q aligned methodologies, that value artful and interpretive uses of methods and reflexivity, you must consider carefully how AI tools can be deployed in a way that remains faithful to these perspectives on the use of methods.

4. Does your use of AI-assisted analysis methods align with the qualitative analysis methods you are using? Both Small-q and Big-Q aligned methods have specific procedures for analysing data and evaluating analysis quality. When you use AI to assist your use of methods you need to make sure it is applied in ways consistent with these procedures.

5. Are the values of your research aligned with the way you are doing qualitative research? As discussed, it is important to avoid situations where the positivist uses of methods are incoherently combined with non-positivist values, and vice versa. For research to be consistent with either a Small-q or Big-Q approach, positivist research values must be aligned with positivist ways of doing qualitative research using AI, and non-positivist research values must be aligned with non-positivist ways of doing qualitative research using AI.

6. Lastly, is your use of AI- assisted data analysis ethical? The use of AI-assisted data analysis can present several challenges that conflict with the ethical standards researchers must uphold (as discussed later).

Exploratory Examples of AI-Assisted Reflexive Content Analysis

We now have an understanding of the two approaches to qualitative research, a model for understanding these approaches with AI defined within it, and questions to explore how the choice of research approach and values influences your use of AI-assisted analysis methods. We now turn our discussion to what these considerations may mean for how analysis is conducted in practice through a series of exploratory examples.

We do not aim to present a detailed explanation of how to conduct an AI-assisted qualitative analysis. The examples are exploratory insofar as they show how different approaches may shape the application of AI-assisted methods; detailed guides on how to use AI to conduct data analysis have been published elsewhere (see, for example, Bano et al., 2023; Bijker et al., 2024; Hitch, 2024; Paoli, 2024; Paoli & Mathis, 2024; Tai et al., 2024; Theelen et al., 2024). Our goal is to demonstrate how thinking about qualitative research approaches and their associated values can change how AI-assisted data analysis is conducted. This is intended to encourage researchers to reflect on how their research values and alignment with either qualitative approach might shape their application of AI-assisted data analysis within the context of their chosen research methods.

We employ reflexive content analysis (Nicmanis, 2024) as our chosen method of data analysis. Reflexive content analysis is a transtheoretical and flexible researcher-oriented method intended for the qualitative description of manifest content into a hierarchical structure of codes, subcategories, and categories based on a research question. Manifest content, refers to the overt surface meaning of data, as opposed to latent content, which encompasses the deeper underlying meanings. Codes are seen as being the closest to the original data while categories are the most abstracted. The analysis is conducted over seven stages: Stage 1 defines the research question, the qualitative approach used, and justifies the use of reflexive content analysis; Stage 2 involves collecting the data and familiarizing the researchers with it; Stage 3 comprises the initial coding of the data; during Stage 4 the codes are revised; Stage 5 is the development of the analysis structure; Stage 6 is the reporting of the analysis structure; and Stage 7 is interpreting and reporting the findings.

Although several types of qualitative content analysis are available, we have chosen reflexive content analysis for this demonstration for three reasons. First, reflexive content analysis is intended solely as a method and thus is transtheoretical and flexible. While this method is generally aligned with a Big-Q approach, this flexibility allows key concepts from reflexive content analysis to be used within Small-q approaches to data analysis. The differences in the use of this method between the two approaches are illustrated in Table 2. Second, reflexive content analysis is intended for the reduction and description of manifest content. The description of the overt surface meanings of data fits well within the limitations of current AI models, which cannot effectively interpret subjective experiences (Hitch, 2024). Third, reflexive content analysis uses reflexivity as an analytical resource in each stage of the analysis process to aid and strengthen the development of findings. In current approaches to AI-assisted data analysis, human choices are needed to generate the prompts used to work with LLMs, make decisions about the analysis, and moderate the analysis (Bijker et al., 2024; Christou, 2023, 2024; Chubb, 2023; Tai et al., 2024; Tschisgale et al., 2023). We see the inclusion of reflexivity and reflections on our choices as researchers in each stage of the analysis as a central mediator between understanding our research values and the use of AI-assisted analysis.

Table 2.

Summary of the Small-Q and Big-Q Approaches to Reflexive Content Analysis

Element	Small-q RCA	Big-Q RCA
Epistemology/ontology	Positivist/realist: Reality is in some way knowable and objective	Non-positivist: Reality is subjective, constructed, or fallible
Conceptual framework	Testable descriptive research questions based on prior evidence and theory	Exploratory descriptive research questions guided by evidence and experience
Methodology	Guides structured and statistical use of methods. Simple qualitative description or numeric description. Reflexivity represents recording of research choices and bias control	Guides artful and interpretive use of methods. Reflexive qualitative description with deep inclusion and acknowledgment of reflexivity throughout the analysis
Method	Structured or statistical description of data based on manifest content	The artful and interpretive description of manifest content drawing on reflexivity
Outcome	Quantitative or simple qualitative description of phenomena (e.g. number of disease symptoms reported, or types of symptoms reported)	Rich descriptions of phenomena and experiences as recounted by the data (e.g. description of the experiential impacts of disease symptoms on participants)

Note. RCA = Reflexive content analysis.

Small-q Aligned AI-Assisted Data Analysis

A Small-q approach to AI-assisted data analysis in reflexive content analysis adheres to positivist values, guiding the structured application of AI and reflexive content analysis. The positivist values of this approach assert that a reality exists and can be understood through research, with the analysis assessed using criteria such as validity and reliability (Lincoln et al., 2011). The application of reflexive content analysis (Nicmanis, 2024) within a Small-q approach focuses on the numeric or simple qualitative description of data to categorise patterns. Given this, AI within a Small-q approach can be seen as a tool that uses complex statistical processes to fully automate or assist analysis, much like how a mass spectrometer in a chemical laboratory is seen as a tool to identify chemical substances.

We provide an example of how AI-assisted analysis could be conducted within a Small-q approach in Figure 3. The AI model can be trained or given instructions to perform stages 3 to 5 of the reflexive content analysis process and the research question to address. The qualitative data is then provided to the AI, which can autonomously inductively code the data, revise codes, create the analysis structure, and tabulate frequencies for each analysis unit with limited human intervention. Given the inherent unpredictability of current LLMs, you may also be able to be run the analysis recursively to produce more stable results (see, for example, Tai et al., 2024).

Figure 3.

An Illustrative Flow-Chart for Conducting a Small-Q Reflexive Content Analysis Using AI-Assistance. Note. AI = “Artificial Intelligence”

This process requires many decisions, including data management, software coding, and prompt engineering, which are all important considerations for qualitative analysis using LLMs (Bijker et al., 2024). As reflexivity is central to reflexive content analysis, from a Small-q perspective, this represents the recording of a researcher’s choices during the AI-assisted analysis process to control for biases. This includes decisions in planning the analysis and considerations such as prompt engineering.

After this process is complete, the quality of the AI-assisted analysis can be assessed based on Small-q principles. Small-q approaches to reflexive content analysis draw on positivist metrics of analysis quality including validity and reliability. These may be assessed by calculating agreement and inter-rater reliability with human coders (for example, Bano et al., 2023; Paoli, 2024; Tai et al., 2024; Theelen et al., 2024). Work by Bijker et al. (2024) also demonstrates that inter-rater reliability may be calculated by comparing multiple instances of a ChatGPT analysis where each instance is an independent coder, highlighting the possibility for AI to act as a second coder during analysis. In the case of Small-q aligned reflexive content analysis, low inter-rater reliability may suggest the analysis structure created is of poor quality and does not fit the data or the analysis process is not reliable. This information can then be used to refine the model iteratively leading to a final analysis structure that effectively addresses the initial research questions.

For example, you may have a large dataset of posts made to social media by hundreds of users who recount their symptoms with a rare disease. Your aim or research question is aligned to Small-q positivist values, and you want to describe the frequency and types of symptoms reported in this textual data to identify potential patterns in this disease. A Small-q use of reflexive content analysis aligned with these values would use an AI system to inductively code the different symptoms and develop an analysis structure that describes the symptoms. You could then check the reliability and validity of a subset of these codes compared to human coding. If necessary, the model or prompts used could then be adjusted. The outcome of the final analysis would be a numeric (e.g., number of symptoms reported) or simple qualitative description (e.g., types of symptoms reported) of the data. For a practical guide and examples of conducting a Small-q-aligned analysis using ChatGPT in the context of qualitative content analysis (rather than reflexive content analysis), see Bijker et al. (2024).

It is important to clarify that applying AI assistance in this manner is neither exploratory nor interpretative; it lacks consideration of the subjectivities of the researcher and the participants represented in the data. It would be inappropriate to use this kind of method to develop a reflexive qualitative description of data or use the outcomes of this analysis to conduct deeper, highly interpretive analyses that construct interpretive themes. For qualitative researchers accustomed to highly interpretive analyses and Big-Q aligned assessments of research quality, such an approach may appear lacking. However, the use of textual data within Small-q approaches has a well-established history in qualitative analysis and holds its own place in the research literature. Therefore, when employing AI assistance in this manner, it must be explicitly identified as a Small-q approach to data analysis to prevent undermining or causing confusion with Big-Q research approaches.

Big-Q Aligned AI-Assisted Data Analysis

The AI-Assisted Continuation of Analysis

Big-Q approaches view reality as subjective, constructed, and fallible, evaluating research quality through metrics such as trustworthiness (Guba & Lincoln, 1982; Lincoln et al., 2011). A Big-Q approach to reflexive content analysis (Nicmanis, 2024) focuses on the reflexive qualitative description of data, leveraging subjectivities to create rich and nuanced descriptions of phenomena and experiences recounted in the manifest meanings of data. Within a Big-Q approach, AI is not seen or used as an autonomous analysis machine that independently uncovers insights from data or conducts analysis, but rather as a tool embedded within a subjective or constructed reality, acting as a fallible assistant that aids the analysis process and reflexivity.

One way of doing a Big-Q aligned reflexive content analysis with AI is the AI-assisted continuation of analysis. While big data analysis is highlighted as a key benefit of AI assistance (Hitch, 2024), the challenge lies in harnessing this capability to assist reflexive content analysis in a manner that leverages the value of reflexivity aligned with Big-Q values. For example, you may want to analyse open-ended questions from a large or nationally representative survey with thousands of respondents. These open-ended questions allowed respondents to provide highly detailed accounts of their experiences with a complex social issue. In approaching this project, you draw on Big-Q aligned research values and want to use reflexive content analysis to reflexively describe the impacts of this social issue on their lives. However, the volume of responses makes it impractical to include them all in a traditional analysis. Figure 4 illustrates how you may be able to include all the responses using the AI-assisted continuation of analysis. To analyze this large dataset of survey responses, the process would begin with a human researcher conducting a reflexive content analysis on a subset of hundreds of responses, selected from a larger dataset containing thousands. The insights gained from this initial analysis can then be used to train or prompt an AI model, enabling it to apply the same analytical structure to the entire dataset.

Figure 4.

An Illustrative Flow Chart for Conducting a Big-Q Aligned AI-Assisted Continuation of Analysis. Note. AI = “Artificial Intelligence”

Importantly, the AI’s role is not to replace human judgment but to enhance the researcher’s analytical capacity. When data diverges from the established structure, the AI flags these anomalies, allowing the researcher to review and incorporate them into their analysis based on their human reflexive insights with subsequent adjustments made to the model as needed. This helps preserve the depth and nuance of human interpretation in the reflexive qualitative descriptions provided by reflexive content analysis aligned with a Big-Q approach.

AI as a Reflexive Collaborator

In the case of current AI models, the second promising application for reflexive content analysis is to engage in dialogues with data, positioning AI as a collaborative tool like a research colleague. While caution should be exercised regarding how current AI systems generate insights, given their inherent interpretive limitations, their ability to process and identify patterns from extensive datasets is invaluable during the later stages of reflexive content analysis, such as developing an analytical structure.

For example, when organising codes into subcategories and categories, which can be intellectually demanding, a conversational AI can serve as a brainstorming partner, challenging categorisations or offering suggestions and thus prompting deeper reflexive thinking. A Big-Q aligned use of AI in this way does not delegate the analytical work to the AI; rather, it enriches the researcher’s reflexive engagement with the data, ensuring that the analysis remains a deeply human-driven endeavour while benefiting from AI’s processing capabilities. A practical example of this method using ChatGPT and reflexive thematic analysis can be seen in the work of Hitch (2024).

Miscellaneous Applications of AI-Assistance

In addition to these uses of AI-assisted analysis, we draw attention to some useful miscellaneous applications of this technology that could be used in either approach. Qualitative research often requires presenting quotes from participants as part of reporting (Elo et al., 2014). In the case of qualitative data gathered from social media, presenting quotes verbatim creates a potential ethical issue, as these could be reverse identified using readily available search engines. In these contexts, the argument has been made that we should instead implement methods to produce composite accounts to protect the identities of participants (Markham, 2012). We could potentially use AI language models to create vignettes (see, for example, Chubb, 2023) or composite accounts from the data to maintain participant anonymity.

Additionally, the manual coding of data required by many analysis methods can be highly error prone. Even though methods like reflexive content analysis incorporate iterative coding loops to minimise such errors (Nicmanis, 2024), inconsistencies in coding and missed data are still likely to occur. An AI-based system could be used to identify uncoded data or flag inconsistencies in how codes have been applied. This could enhance the manual analysis process without using AI to conduct the analysis.

Ethical Limitations of AI-Assisted Data Analysis

Even when integrated in a manner aligned with either of the two qualitative research approaches, AI-assisted data analysis comes with serious ethical implications that must be considered. While a full discussion is beyond the scope of this paper, as there are no formal ethical guidelines for AI-assisted analysis, we must highlight key emerging challenges.

The use of commercially available AI tools like ChatGPT in qualitative research currently necessitates sharing sensitive data with third-party, privately owned platforms that can profit from the data, raising substantial issues with data ownership and the rights of participants (Davison et al., 2024; Hitch, 2024; Marshall & Naff, 2024). In the case of qualitative research, often sensitive and personal information is obtained through interviews and the observation of interactions. Sharing this data with third parties not only compromises the confidentiality promised to participants but may also violate ethical standards and legal regulations concerning data protection.

Additionally, the way these technologies operate, including their potential to fabricate information, misunderstand experiences, or produce highly biased analyses, raises substantial ethical concerns (Christou, 2023; Chubb, 2023; Marshall & Naff, 2024; Roberts et al., 2024; Tai et al., 2024). Researchers using a Small-q approach may use AI assistance to enhance objectivity and validity in their analysis (Hitch, 2024); however, AI-generated outputs can reflect bias in how these models were trained. For example, LLMs trained on Western data might misinterpret cultural nuances in interviews with minority populations and reinforce WEIRD (Western, educated, industrial, rich, and democratic) biases (Henrich et al., 2010). LLMs also have the tendency to hallucinate or fabricate information in outputs due to their underlying mechanisms that prioritise probabilistic accuracy over factual correctness (Alkaissi & McFarlane, 2023; Emsley, 2023).

Researchers using a Big-Q approach, in contrast, may attempt to use AI to help capture rich nuanced insights from their data. In the case of current AI models like ChatGPT, they lack creativity, personal style, and the lived experience to interpret data (Hitch, 2024). Instead, these models produce analyses by predicting the most likely response based on probabilities from their training data, inherently prioritising central tendencies (the mean) and overlooking less frequent or unexpected insights (the tail-end), which is a substantial limitation for the depth required in Big-Q research (Bender et al., 2021; Vaswani et al., 2017). For example, an LLM trained on a mainstream internet dataset might produce predictable surface-level interpretations of complex phenomena, thereby smoothing out the richness and diversity that Big-Q aligned researchers aim to capture. Therefore, AI-assisted data analysis used uncritically may produce cookie-cutter research that only serves to reinforce existing explanations of phenomena and risks perpetuating bias or fabricated accounts.

Future Directions

As AI becomes increasingly integrated into qualitative analysis, it is vital to ensure its use is correct, transparent, and aligned with the established approaches to understanding qualitative research. At a minimum, researchers must clearly identify their values, research approach, and use of methods. There are currently no specific ethical guidelines for AI-assisted qualitative data analysis, and general guidelines on the use of AI are sparse (Davison et al., 2024). There is a need for clear reporting standards, ethical guidelines, and transparency in the use of AI-assisted data analysis so that these technologies are not used in ways that undermine the integrity of research.

While we have discussed theoretical challenges arising from the use of AI-assisted data analysis methods, much work remains to be done. Our exploratory examples have focused only on how AI can be used to assist one type of content analysis, highlighting the need for further research to explore how AI can support a broader range of qualitative methods. There is also a need to develop AI tools and models specifically designed to support qualitative analysis that address the limitations and ethical challenges of current models. This should include ways of assessing the quality of AI-assisted analyses that respect the values of the two approaches to qualitative research. In addition, we believe the models presented in this paper are valuable for clarifying qualitative analysis concepts outside the field of AI-assisted data analysis.

Conclusion

In writing this paper, we do not advocate that all qualitative researchers adopt AI-assisted data analysis methods. Some methods of qualitative research will be fundamentally incompatible with AI. Instead, we urge researchers who choose to use this technology to thoughtfully integrate AI-assisted data analysis methods with their research values, thereby harnessing the potential benefits of AI while upholding the values and traditions of the two approaches to qualitative research: Small-q and Big-Q. Failing to do so risks misaligning qualitative research values and methods, leading to potentially inconsistent analyses and perpetuating misunderstandings about qualitative research. Therefore, those using AI-assisted data analysis methods should always be clear about not just their methods but also their research values and qualitative research approach.

Footnotes

Acknowledgments

We would like to express our sincere gratitude to Mitchell’s supervisors,Anna Chur-Hansen and Melissa Oxlad,and Harry’s supervisor,Rajeev Kamineni,Manjula Dissanayake,and Noel Lindsay,for their invaluable feedback and guidance. We are both deeply grateful for our respective supervisors continuing support and encouragement. We also would like to thank Nicholas Pritchard for his feedback on our descriptions of artificial intelligence and Georgia Breakey,Erin Barrera,and Olivia Arcangeli for their time and effort helping draft this paper.

ORCID iDs

Mitchell Nicmanis

Harry Spurrier

Funding

The author(s) received no financial support for the research,authorship,and/or publication of this article.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research,authorship,and/or publication of this article.

References

Alkaissi

McFarlane

S. I.

(2023). Artificial hallucinations in ChatGPT: Implications in scientific writing. Cureus, 15(2), Article e35179. https://doi.org/10.7759/cureus.35179

Bano

Zowghi

Whittle

(2023). AI and human reasoning: Qualitative research in the age of large language models. AI Ethics Journal, 3(1), 1–15. https://doi.org/10.47289/AIEJ20240122

Bender

E. M.

Gebru

McMillan-Major

Shmitchell

(2021). On the dangers of stochastic parrots. In Proceedings of the 2021 ACM conference on fairness, accountability, and transparency (pp. 610–623). Association for Computing Machinery. https://doi.org/10.1145/3442188.3445922

Bijker

Merkouris

S. S.

Dowling

N. A.

Rodda

S. N.

(2024). ChatGPT for automated qualitative research: Content analysis. Journal of Medical Internet Research, 26, Article e59050. https://doi.org/10.2196/59050

Bokhove

Downey

(2018). Automated generation of ‘good enough’ transcripts as a first step to transcription of audio-recorded data. Methodological Innovations, 11(2), Article 205979911879074. https://doi.org/10.1177/2059799118790743

Braun

Clarke

(2019). Reflecting on reflexive thematic analysis. Qualitative Research in Sport, Exercise and Health, 11(4), 589–597. https://doi.org/10.1080/2159676X.2019.1628806

Braun

Clarke

(2021). One size fits all? What counts as quality practice in (reflexive) thematic analysis? Qualitative Research in Psychology, 18(3), 328–352. https://doi.org/10.1080/14780887.2020.1769238

Braun

Clarke

(2022a). Conceptual and design thinking for thematic analysis. Qualitative Psychology, 9(1), 3–26. https://doi.org/10.1037/qup0000196

Braun

Clarke

(2022b). Thematic analysis: A practical guide. Sage.

10.

Braun

Clarke

(2023). Toward good practice in thematic analysis: Avoiding common problems and be(com)ing a knowing researcher. International Journal of Transgender Health, 24(1), 1–6. https://doi.org/10.1080/26895269.2022.2129597

11.

Brown

T. B.

Mann

Ryder

Subbiah

Kaplan

Dhariwal

Neelakantan

Shyam

Sastry

Askell

Agarwal

Herbert-Voss

Krueger

Henighan

Child

Ramesh

Ziegler

D. M.

Winter

Amodei

(2020). Language models are few-shot learners. In Larochelle

Ranzato

Hadsell

Balcan

M. F.

Lin

(Eds.), Advances in neural information processing systems 33. Curran Associates.

12.

Bush

A. A.

Amechi

Persky

(2020). An exploration of parmacy education researchers’ perceptions and experiences conducting qualitative research. American Journal of Pharmaceutical Education, 84(3), ajpe7129. https://doi.org/10.5688/ajpe7129

13.

Carter

S. M.

Little

(2007). Justifying knowledge, justifying method, taking action: Epistemologies, methodologies, and methods in qualitative research. Qualitative Health Research, 17(10), 1316–1328. https://doi.org/10.1177/1049732307306927

14.

Chamberlain

(2000). Methodolatry and qualitative health research. Journal of Health Psychology, 5(3), 285–296. https://doi.org/10.1177/135910530000500306

15.

Chamberlain

(2015). Epistemology and qualitative research. In Rohleder

Lyons

A. C.

(Eds.), Qualitative research in clinical and health psychology (pp. 9–29). Palgrave Macmillan.

16.

Christou

(2023). Ηow to use artificial intelligence (AI) as a resource, methodological and analysis tool in qualitative research? Qualitative Report, 28(7), 1968–1980. https://doi.org/10.46743/2160-3715/2023.6406

17.

Christou

(2024). Thematic analysis through artificial intelligence (AI). Qualitative Report, 29(2), 560–576. https://doi.org/10.46743/2160-3715/2024.7046

18.

Chubb

L. A.

(2023). Me and the machines: Possibilities and pitfalls of using artificial intelligence for qualitative data analysis. International Journal of Qualitative Methods, 22, Article 16094069231193593. https://doi.org/10.1177/16094069231193593

19.

Chung-Lee

Lapum

(2024). Resisting the robotic – moving fluidly towards soulful qualitative analysis. International Journal of Qualitative Methods, 23, Article 16094069241247475. https://doi.org/10.1177/16094069241247475

20.

Collins

C. S.

Stockton

(2022). The theater of qualitative research: The role of the researcher/actor. International Journal of Qualitative Methods, 21, Article 160940692211031. https://doi.org/10.1177/16094069221103109

21.

Collins

C. S.

Stockton

C. M.

(2018). The central role of theory in qualitative research. International Journal of Qualitative Methods, 17(1), Article 160940691879747. https://doi.org/10.1177/1609406918797475

22.

Davison

R. M.

Chughtai

Nielsen

Marabelli

Iannacci

van Offenbeek

Tarafdar

Trenz

Techatassanasoontorn

A. A.

Díaz Andrade

Panteli

(2024). The ethics of using generative AI for qualitative data analysis. Information Systems Journal, 34(5), 1433–1439. https://doi.org/10.1111/isj.12504

23.

Dwivedi

Y. K.

Kshetri

Hughes

Slade

E. L.

Jeyaraj

Kar

A. K.

Baabdullah

A. M.

Koohang

Raghavan

Ahuja

Albanna

Albashrawi

M. A.

Al-Busaidi

A. S.

Balakrishnan

Barlette

Basu

Bose

Brooks

Buhalis

Wirtz

(2023). “so what if ChatGPT wrote it?” Multidisciplinary perspectives on opportunities, challenges and implications of generative conversational AI for research, practice and policy. International Journal of Information Management, 71, Article 102642. https://doi.org/10.1016/j.ijinfomgt.2023.102642

24.

Elo

Kääriäinen

Kanste

Pölkki

Utriainen

Kyngäs

(2014). Qualitative content analysis. Sage Open, 4(1), Article 215824401452263. https://doi.org/10.1177/2158244014522633

25.

Emsley

(2023). Chatgpt: These are not hallucinations – they’re fabrications and falsifications. Schizophrenia, 9(1), 52. https://doi.org/10.1038/s41537-023-00379-4

26.

Franzosi

(1990). Computer-assisted coding of textual data. Sociological Methods & Research, 19(2), 225–257. https://doi.org/10.1177/0049124190019002004

27.

Guba

E. G.

Lincoln

Y. S.

(1982). Epistemological and methodological bases of naturalistic inquiry. Educational Technology Research & Development, 30(4), 233–252. https://doi.org/10.1007/BF02765185

28.

Harwood

T. G.

Garry

(2003). An overview of content analysis. The Marketing Review, 3(4), 479–498. https://doi.org/10.1362/146934703771910080

29.

Henrich

Heine

S. J.

Norenzayan

(2010). Most people are not WEIRD. Nature, 466(7302), 29. https://doi.org/10.1038/466029a

30.

Hitch

(2024). Artificial intelligence augmented qualitative analysis: The way of the future? Qualitative Health Research, 34(7), 595–606. https://doi.org/10.1177/10497323231217392

31.

Jovanović

(2011). Toward a social history of qualitative research. History of the Human Sciences, 24(2), 1–27. https://doi.org/10.1177/0952695111399334

32.

Kidder

L. H.

Fine

(1987). Qualitative and quantitative methods: When stories converge. New Directions for Program Evaluation, 1987(35), 57–75. https://doi.org/10.1002/ev.1459

33.

Lee

L. W.

Dabirian

McCarthy

I. P.

Kietzmann

(2020). Making sense of text: Artificial intelligence-enabled content analysis. European Journal of Marketing, 54(3), 615–644. https://doi.org/10.1108/EJM-02-2019-0219

34.

Lincoln

Y. S.

Lynham

S. A.

Guba

E. G.

(2011). Paradigmatic controversies, contradicitions, and emerging confluences, revisted. In Denzin

N. K.

Lincoln

Y. S.

(Eds.), The Sage handbook of qualitative research (4th ed., pp. 97–129). Sage.

35.

Markham

(2012). Fabrication as ethical practice. Information, Communication & Society, 15(3), 334–353. https://doi.org/10.1080/1369118X.2011.641993

36.

Marshall

D. T.

Naff

D. B.

(2024). The ethics of using artificial intelligence in qualitative research. Journal of Empirical Research on Human Research Ethics, 19(3), 92–102. https://doi.org/10.1177/15562646241262659

37.

McCosker

L. K.

Sendall

M. C.

(2015). Undergraduate public health students and perceptions of qualitative research: A trajectory of understanding from airy fairy to a valuable tool. Learning and Teaching, 8(1), 19–36. https://doi.org/10.7459/lt/8.1.03

38.

Natow

R. S.

(2022). Policy actors’ perceptions of qualitative research in policymaking: The case of higher education rulemaking in the United States. Evidence & Policy, 18(1), 109–126. https://doi.org/10.1332/174426420X16047229138089

39.

Nicmanis

(2024). Reflexive content analysis: An approach to qualitative data analysis, reduction, and description. International Journal of Qualitative Methods, 23, Article 16094069241236603. https://doi.org/10.1177/16094069241236603

40.

O’Connor

Joffe

(2020). Intercoder reliability in qualitative research: Debates and practical guidelines. International Journal of Qualitative Methods, 19, Article 160940691989922. https://doi.org/10.1177/1609406919899220

41.

Paoli

S. de

(2024). Performing an inductive thematic analysis of semi-structured interviews with a large language Model: An exploration and provocation on the limits of the approach. Social Science Computer Review, 42(4), 997–1019. https://doi.org/10.1177/08944393231220483

42.

Paoli

S. de

Mathis

W. S.

(2024). Reflections on inductive thematic saturation as a potential metric for measuring the validity of an inductive thematic analysis with LLMs. Quality and Quantity. Advance online publication. https://doi.org/10.1007/s11135-024-01950-6

43.

Pascale

C. M.

(2011). Cartographies of knowledge: Exploring qualitative epistemologies. Sage.

44.

Paulus

T. M.

Marone

(2024). “In minutes instead of weeks”: Discursive constructions of generative AI and qualitative data analysis. Qualitative Inquiry, Article 10778004241250065. Advance online publication. https://doi.org/10.1177/10778004241250065

45.

Povee

Roberts

L. D.

(2014). Qualitative research in psychology: Attitudes of psychology students and academic staff. Australian Journal of Psychology, 66(1), 28–37. https://doi.org/10.1111/ajpy.12031

46.

Roberts

Baker

Andrew

(2024). Artificial intelligence and qualitative research: The promise and perils of large language model (LLM) ‘assistance.' Critical Perspectives on Accounting, 99 Article 102722. https://doi.org/10.1016/j.cpa.2024.102722

47.

Tai

R. H.

Bentley

L. R.

Xia

Sitt

J. M.

Fankhauser

S. C.

Chicas-Mosier

A. M.

Monteith

B. G.

(2024). An examination of the use of large language models to aid analysis of textual data. International Journal of Qualitative Methods, 23, Article 16094069241231168. https://doi.org/10.1177/16094069241231168

48.

Theelen

Vreuls

Rutten

(2024). Doing research with help from ChatGPT: Promising examples for coding and inter-rater reliability. International Journal of Technology in Education, 7(1), 1–18. https://doi.org/10.46328/ijte.537

49.

Tschisgale

Wulff

Kubsch

(2023). Integrating artificial intelligence-based methods into qualitative research in physics education research: A case for computational grounded theory. Physical Review Physics Education Research, 19(2), Article 020123. https://doi.org/10.1103/PhysRevPhysEducRes.19.020123

50.

Vaswani

Shazeer

Parmar

Uszkoreit

Jones

Gomez

A. N.

Kaiser

\. u.

Polosukhin

(2017). Attention is all you need. In Guyon

Von Luxburg

Bengio

Wallach

Fergus

Vishwanathan

Garnett

(Eds.), 31st Conference on Neural Information Processing System (30). Curran Associates. https://proceedings.neurips.cc/paper_files/paper/2017/file/3f5ee243547dee91fbd053c1c4a845aa-Paper.pdf.

51.

Webb

(1999). Analysing qualitative data: Computerized and other approaches. Journal of Advanced Nursing, 29(2), 323–330. https://doi.org/10.1046/j.1365-2648.1999.00892.x

Getting Started with Artificial Intelligence Assisted Qualitative Analysis: An Introductory Guide to Qualitative Research Approaches with Exploratory Examples from Reflexive Content Analysis

Abstract

Keywords

Introduction

Paper Outline

Background

The Two Approaches to Qualitative Research

Current Work in AI-Assisted Qualitative Analysis

The Relevancy of the Two Qualitative Approaches for AI-Assisted Analysis Methods

An Approach-Based Model of Qualitative Research

Research Values

Epistemology and Ontology

Methodology

Conceptual Framework

Doing Qualitative Research

Methods

Outcomes

Defining AI Within the Approach-Based Model of Qualitative Research

Questions to Integrate AI-Assisted Data Analysis

Exploratory Examples of AI-Assisted Reflexive Content Analysis

Small-q Aligned AI-Assisted Data Analysis

Big-Q Aligned AI-Assisted Data Analysis

The AI-Assisted Continuation of Analysis

AI as a Reflexive Collaborator

Miscellaneous Applications of AI-Assistance

Ethical Limitations of AI-Assisted Data Analysis

Future Directions

Conclusion

Footnotes

Acknowledgments

ORCID iDs

Funding

Declaration of Conflicting Interests

References