Research Article

International Journal of Qualitative Methods

Volume 24: 1–18
Thematic Analysis and Artificial © The Author(s) 2025
DOI: 10.1177/16094069251333886
Intelligence: A Step-by-Step Process for journals.sagepub.com/home/ijq

Using ChatGPT in Thematic Analysis

Muhammad Naeem1, Tracy Smith1, and Lorna Thomas1

Abstract
This study sets out how to use generative artificial intelligence (AI) in the six steps of systematic thematic analysis. It leverages AI
to address the limitations of traditional thematic analysis. This paper developed prompts (inputs) for ChatGPT (a generative AI
chatbot based on a large language model) that are based on many researchers’ discussions and criticisms of qualitative data
analysis. The contributions of this paper are twofold. First, it addresses a critical research gap by showcasing ChatGPT prompts
for each step of the six steps of systematic thematic analysis, which also addresses researcher training in thematic analysis.
Second, it contributes to the development of input to train AI in thematic analysis, including a description of how to familiarize
an AI system with the context of a research study and the researcher’s methodological and theoretical considerations; this
approach helps to reduce human bias and improves accountability and transparency in thematic analysis.

Keyword
thematic analysis, themes codes, artificial intelligence, ChatGPT prompts, qualitative research

Introduction Recent studies have explored the use of Generative AI

(GenAI) in qualitative research. For example, De Paoli, 2024,
Researchers have begun to explore the use of Artificial Intelli- 2024b compared the results of manual thematic analysis with
gence (AI) in research by, for example, conducting literature the results of GenAI-based thematic analysis. De Paoli, 2024,
reviews (Švab et al., 2023), literature synthesis (Semrl et al., 2024b used open access data sets and semi-structured inter-
2023), data management and analysis (Currie et al., 2023b), data views that had been analysed by other researchers. De Paoli,
interpretation (Laios et al., 2023), and considerations of academic 2024, 2024b used GenAI to carry out initial coding of data sets
integrity and workflow efficiency (Currie et al., 2023a). The use and to generate themes, and compared the themes produced by
of AI applications in scientific writing has also been explored GenAI-based thematic analysis with those derived from
(Huang & Tan, 2023; Salimi & Saheb, 2023). AI can be used to manual thematic analysis. De Paoli (2024a) found that GenAI
improve the articulation, coherence and clarity of academic inferred most of the main themes identified by manual coding.
writing for writers whose first language is not English (Meyer However, these studies did not include the familiarization
et al., 2023; Salimi & Saheb, 2023). The use of AI in qualitative phase of thematic analysis, which would have allowed the
research expedites the data analytical process (Prescott et al.,
2024), but it also raises issues of conceptuality, ethics and
transparency (Meyer et al., 2023; Salimi & Saheb, 2023). 1
Christ Church Business School, Canterbury Christ Church University,
Moreover, Naeem et al. (2025) used ChatGPT to conduct sys- Canterbury, UK
tematic thematic analysis in their study, explaining the use of AI
Corresponding Author:
as follows: “AI-assisted technology was employed specifically to Muhammad Naeem, Christ Church Business School, Canterbury Christ
select keywords and quotations from the data” (Naeem et al., Church University, North Holmes Road, Canterbury CT1 1QU, UK.
2025, p., 12). Email: [email protected]

Creative Commons CC BY: This article is distributed under the terms of the Creative Commons Attribution 4.0 License (https://
creativecommons.org/licenses/by/4.0/) which permits any use, reproduction and distribution of the work without further
permission provided the original work is attributed as specified on the SAGE and Open Access pages (https://us.sagepub.com/en-us/nam/
open-access-at-sage).
2 International Journal of Qualitative Methods

generative AI to engage with the data and overall research and themes, and (6) the development of a conceptual model (see
context. To overcome this issue, Lee et al. (2024) used AI to Figure 1). The significant difference with a systematic review is
transcribe interviews to familiarize an AI system with raw that through this process, researchers explore and identify
data. However, they only used AI to perform three of the steps themes based on required data, considering research gaps,
accounted for in Braun and Clarke’s (2006) six steps to objectives, and questions, instead of finding common themes
thematic analysis. Prescott et al. (2024) qualitatively tested the from the data. Therefore, this systematic thematic analysis al-
efficacy and efficiency of thematic analysis codes generated lows researchers to create themes based on identified quotations
by humans and GenAI. Prescott’s et al. (2024) findings in- and keywords, grouping them into different categories known
dicated that human-generated codes are more reliable than AI- as codes (Naeem et al., 2023). These codes are based on relevant
generated codes. These studies have three limitations. The first quotations and keywords. Subsequently, these codes are
is that they did not follow all six steps of Braun and Clarke’s grouped according to research questions and objectives,
(2006) approach to thematic analysis. The second limitation is forming what is called a theme. Thus, through systematic
that the studies did not provide the AI system with contextual thematic analysis, themes are derived from the data based on
information about the research, such as the aim and the research research gaps, theoretical underpinnings, philosophical under-
questions. The third limitation was that the AI system was not pinnings, and research questions, whereas in traditional the-
informed about the researchers’ methodological considerations matic analysis, themes are based on commonalities in the data
for each step of thematic analysis. (Naeem, 2025; Naeem et al., 2025; Naeem & Ozuem, 2025).
This paper addresses researcher training in thematic Building on the groundwork of the above-mentioned
analysis, and the development of input to train AI in thematic studies and addressing their limitations, the current paper
analysis. The paper explores each step involved in thematic incorporates their insights into ChatGPT prompts to enable the
analysis, and considers the methodology underpinning each chatbot to conduct systematic thematic analysis. (Naeem et al.,
step. In this sense, it addresses Christou’s plea (2023a, p. 567) 2023). The results of a systematic thematic analysis conducted
when he noted “I strongly encourage, particularly novice by ChatGPT are compared with the corresponding manual
researchers, to first equip themselves with the fundamentals of outcomes of Naeem et al. (2024a). This paper was selected as a
thematic analysis and how to use AI tools in their thematic case study since it uses the same systematic thematic analysis
analysis”. In addition, the paper describes how researchers can process in their study. The primary data set used by Naeem
familiarize an AI system with the context of research, in- et al. (2024a) was used as the primary data set for this paper,
cluding aspects such as the research aim, and the researcher’s and this has facilitated a comparison between the outcomes of
methodological considerations. Christou (2023b) argues that, a manual thematic analysis of Naeem et al., 2024b to compare
to fully utilize AI to generate knowledge, researchers must with a ChatGPT-generated thematic analysis. The research
carefully consider input methods. This paper describes various context, methodology, and philosophical and theoretical un-
input (prompts) given to ChatGPT to complete the six steps of derpinnings of Naeem et al., 2024b study was also part to
thematic analysis . and it provides the rationale for these develop ChatGPT prompt in this methodology paper.
prompts. ChatGPT is a GenAI chatbot based on a large
language model. Textual inputs to ChatGPT are called How can ChatGPT be Used in the Process of
“prompts”. These prompts can be thought of as a conversation
Thematic Analysis?
in which a person asks a question, gives instructions or ex-
plains the context of a situation to their assistant, so that the The development of an AI-based toolkit for systematic the-
assistant can help them perform a task. matic analysis was first introduced by (Naeem et al., 2023).
Braun and Clarke (2006) introduced a six-step approach to This toolkit not only facilitates the application of AI in sys-
thematic analysis comprising of familiarization with the data, tematic thematic analysis, but also offers researchers a well-
generating initial codes, searching for themes, reviewing structured and rigorous approach to use the developed
themes, defining and naming themes and producing a report. prompts for each stage. These prompts have been developed
Additionally, Braun and Clarke (2022) suggested considering following consideration of the different points of view that
reflexivity, theoretical frameworks, data contextualization, and need to be considered at each stage. The paper sets out how to
theme development in the process of thematic analysis. They use ChatGPT for thematic analysis by providing step-by-step
also indicated that thematic analysis encompasses a family of guidance, and a conceptual model (see Figure 2).
methods rather than a singular approach, with no standardized Figure 2 outlines a comprehensive approach to using
version. This paper follows a modified version of the six steps ChatGPT for systematic thematic analysis. A more detailed
introduced by Naeem et al. (2023), which is known as sys- explanation of each step is provided below.
tematic thematic analysis. Naeem et al.’s (2023) systematic
thematic analysis also comprises six steps: (1) transcription, · Step 1: Familiarization, and Selection of Quotations: The
familiarization with data, and the selection of quotations, (2) the first stage of systematic thematic analysis involves fa-
selection of keywords, (3) coding, (4) theme development, (5) miliarizing ChatGPT with the data, research context and
conceptualization through the interpretation of keywords, codes the theoretical, methodological and philosophical
Naeem et al. 3

Figure 1. Systematic Thematic Analysis Process (Naeem et al., 2023, p. 4).

underpinnings of the research including the research codes. The codes should reflect the meaning of the
context. It is also necessary at this stage to familiarize grouped quotations and keywords in order to ad-
ChatGPT with the six steps of systematic thematic analysis dress research questions. Codes should be selected
introduced by Naeem et al., (2023). Researchers then need on the basis of the 6 Rs (reciprocal, recognizable,
to ask ChatGPT, using the instructions provided in Table 1, responsive, resourceful). During this stage it is
to confirm that it understands the context, and is ready to necessary to instruct AI to suggest code names
perform a systematic thematic analysis. based on the instructions in Table 3.
· Step 2 Selection of Keywords: Keywords are rich · Step 4 Theme Development: This stage involves
words or phrases that are selected on the basis of the organising the codes into categories on the basis of
6 Rs (realness, richness, repetition, rationale, rep- their inter-relationships. This is achieved by con-
artee and regal). Keywords ensure the robustness sidering selected theory/theories to label the cate-
and relevance of extracted data (Naeem et al., gories as themes. The codes should reflect the
2023). This stage involves asking AI to select meaning of the grouped quotations and keywords in
keywords form the data on the basis of the prompts order to address research questions. Themes should
given in Table 2. be selected on the basis of the 6 Rs (). The researcher
· Step 3 Coding: Keywords and quotations are must guide the AI by providing the keywords, codes,
considered to label phrases or words to produce research context, research aims and theoretical
4 International Journal of Qualitative Methods

Figure 2. Using ChatGPT for Systematic Thematic Analysis: A Step-by-Step Process Using Artificial Intelligence in Qualitative Research.

underpinnings to ensure that themes can be developed and go apps to shop in supermarkets in England. Their study,
(see Table 4). which adopted a constructivist ethnographic approach, was
· Step 5 Conceptualization: Conceptualization in- underpinned by flow- and affordance theory. They introduced
volves interpreting codes and themes in a coherent the Affordances Flow Funnel concept to conceptualize cus-
manner to help readers understand, categorize, and tomers’ experiences of shopping using scan and go apps.
communicate a meaningful representation of the Systematic thematic analysis was conducted on online reviews
subject at hand (Naeem et al., 2023). The purpose of of 10 scan and go apps.
conceptualization is to define and clarify new con-
cepts derived either from themes and codes or from A Comparison Between Outcomes of
theory. This involves refining codes, defining con-
nections, identifying similarities and differences
GenAI-Based and Manual Systematic
between the themes and codes, and creating a concept Thematic Analysis
supported by existing theories. The researcher Step 1: Familiarization With Background Information
therefore needs to instruct the AI to conceptualise the
and Transcript for Quotation Selection
various codes and themes using selected keywords
and the theoretical underpinning of the (see Table 6). The first step of systematic thematic analysis involves the
· Step 6 Development of Conceptual Model: The final researcher familiarizing themself with the data/transcript
step involves synthesizing and presenting all of the while considering the associated theoretical, contextual and
concepts into a coherent model or framework by philosophical underpinnings (Davidson, 2018; Naeem et al.,
identifying the relationships between the concepts as 2023, 2024b; Nowell et al., 2017; Slembrouck, 2007; Thorne,
arrows and boxes. This involves establishing the sig- 2000). This approach is necessary because reading a transcript
nificance of the relationship between these concepts to is an interpretative task (Eldh et al., 2020), so the researcher
create solutions to the research question, and to provide must consider the research aim, objectives and research
a theoretical contribution. During this stage, the re- questions (Du Bois, 1991; Ochs, 1979, 1999). Additionally,
searcher can ask the AI to present these concepts in a researchers need to consider theoretical and philosophical
structured way using the prompts developed in Table 6. positions to interpret transcripts effectively (Naeem et al.,
2023), which can help reduce research biases introduced by
the transcriber’s perspective (Duranti, 2006; Jaffe, 2000,
This paper presents a case study of Naeem et al.’s (2024a) 2007; Mondada, 2007) when selecting relevant quotations
research to compare each step of systematic thematic analysis (Naeem et al., 2023).
and AI based thematic analysis outcomes. Naeem et al. Since the selected quotations can either reinforce or
(2024a) explored the constraints and enablers of using scan challenge existing knowledge, having some theoretical
Naeem et al. 5

Table 1. ChatGPT Prompts in First Step of Systematic Thematic Analysis

Prompts to familiarise ChatGPT with the Prompts to enable ChatGPT to select

research quotations from the transcript Rationale for prompts

This research analyses primary data to achieve Taking into account the following collective Naeem et al. (2023, 2024a, 2024b) argued
the research aim by answering the following research aim, objectives, methodology, the that researchers need to familiarize
research questions. The primary aim of this theoretical framework and philosophical themselves with the theoretical,
study is [research purpose]. Based on underpinnings, please identify relevant contextual and philosophical backgrounds
[theoretical underpinning of research], this questions to extract excerpts from that of their research in order to justify their
study follows [a brief mention of the data file that answer the research question. choice of relevant quotations. The need
methodology employed in the study (e.g., The selection process must be guided by for familiarization in quotation selection
constructivist ethnographic approach, the specific context of the study, including was also suggested by other researchers
exploratory research, inductive approach, its philosophical foundations, research aim, (Davidson, 2018; Duranti, 2006; Jaffe,
qualitative research method)]. research objectives, research questions, as 2000; Mondada, 2007; Nowell et al., 2017;
Research aim & objectives: [Paste the research well as its methodological and theoretical Slembrouck, 2007; Thorne, 2000).
aim, objective] approaches.
Research question: [Paste the main research • Research aim & objectives: [Insert the
question (and optionally information on the specific aim and objectives of the research]
research problem]. • Methodology: [Insert the methodology
Context of the research: used in the study]
• Data type: [Paste the qualitative data from app • Theoretical Framework: [Insert the
reviews, semi-structured interviews, focus theoretical framework]
group discussions, including transcripts] • Research Philosophy: [Insert the
• Participating and selected Organizations: philosophical approach guiding the study]
[Number of participants, background of • Research Question: [Insert the specific
participants, name of industry, background of research question(s) the study aims to
selected organisations] answer]
Philosophical underpinnings: Data file: [Paste the data set, or attach a data
• [Include an explanation of the epistemology by file, from which quotations will be selected]
copying and pasting the methodology section
of your research]
Theoretical framework:
• Theoretical Framework: [Explain theoretical
frameworks used to guide the research]
Data file: [attach the data file to be thematically
analysed]
Prior to reading the manuscript, please review
the above background material to familiarize
yourself with the research problem, research
question and research methodology and
context. I will ask you to undertake
systematic thematic analysis involving: the
selection of quotations, selection of
keywords, development of codes,
development of themes, and the
conceptualization and development of a
conceptual framework using the following six
steps to systematic thematic analysis
introduced by Naeem et al., (2023). [ask
ChatGPT to write a summary of Naeem et al.
s (2023) systematic thematic analysis,
including the 6 Rs to select keywords, the
6 Rs of coding and the 4 Rs of themes].
(ChatGPT prompt supported/developed)
Source: ChatGPT supported/developed table.
6 International Journal of Qualitative Methods

Table 2. Second step – ChatGPT Prompt, and Keywords Selected by Humans and ChatGPT
Keywords selected Keywords manually selected by Naeem et al.
ChatGPT prompt to perform second step Justification of prompt by ChatGPT (2024a)

Based on the systematic thematic analysis Various authors have identified the following “Shopping • “Not my job, job-eating app, extra burden,
process introduced by Naeem et al. (2023) considerations when selecting keywords in experience” lack of motivation, not learning, do not use,
and the crucial role of keyword selection in thematic analysis. “Technical issues” not tech savvy, challenge”.
systematic thematic analysis, please consider Core idea and experience: Select keywords “Savings” • “Quick process, lost earnings, happy to try,
the following information to aid your that represent participants’ major ideas and “Digital coupons” time wasting, various benefits, installation
selection of appropriate keywords. capture the major chunks of participants’ “ExpressPay” issues, no queue, job loss, shopping by own
Please make sure that the keywords reflect the experiences (Naeem et al., 2023, 2024b). “Location app, fear to lose mobile”.
participants’ points of view and insights to Data saturation and integration: Use helpful detection” • “Enjoyable in smartphones, fell down, free
directly answer the research question and keywords to integrate various data points “Pickup” internet, hard to connect everywhere,
understand the research phenomenon under and various participant perspectives (Morse, “Navigation customer support, non-existent, free trial
study. Please utilize the 6 Rs framework – 2015; Naeem et al., 2023). problems” services, sometimes deduct charges,
Realness, Richness, Repetition, Rationale, Organic selection: Ensure the selected words “Screen brightness” shopping life easy, scanning burden, use of
Repartee and Regal – to guide your selection. and phrases are meaningful to build narrative “Search own device, battery drain”.
Please consider the following information on the bases of different keywords captured functionality” • “Quickly scan items, barcode scanning issue,
about the research in your selection of from participant data (Naeem et al., 2023). “Order processing” relevant deals, no added value, love the
keywords. Linking theory and data: Keywords should be “Pop-up windows” shopping, many bugs, convenient queue for
Research question: [Insert research question] related to the research problem, research “Advertising” QR scan, time saving, stuck in the middle,
Phenomenon under study: [Describe the questions, theories, methodology and “App review mutual list, privacy”.
research aim and research context] research problem (Charma, 2014; Glaser & request” • “Too long to load, complicated app design,
Data file: [Paste the data set or attach the data Strauss, 2009; Naeem et al., 2023; Richards, “In-store pickup” not for old and uneducated people, use
file] 2015; Maxwell, 2013). “Product mobile banking in my brother’s home, text is
In selecting keywords, please ensure they: availability” not readable for me”.
• are meaningful and help to answer the “Sam’s Club” • “Garbage app, not working, looking to
research question, and “Warehouse deals” increase buying, data could be lost, charge
• act as analytical anchors, that is, they are “Shopping list” without buying, quit it, not offering value, lack
helpful in building a narrative based on the “Fingerprint login” of information/could not find, doesn’t sync in
participants’ own words to answer the “Grocery delivery” real time”.
research question. (ChatGPT supported/ “Shopping cart”
developed prompt). “Authentication”
“Location service”
“Members sign up”
“Easy process”
“Repeat shopping”
“Reliability issues”
“Card scan feature”
“Nutritional facts”
“Photo zoom”
“Window resizing”
“Google Pixel Fold”
“Sign-in issues”
“Loading speed”
“Promo page”
“Holiday deals”
“Glitchy app”
“Cart
disappearance”
“In-stock items”
“Search
functionality”
“Back navigation”
“Shop tab”
“Blank screen”
“Item search”
“App loop”
“Screen refresh”
“App reliability”
“Purchase history”
“Account details”
“Slow app”
“Page crashes”
“Cart view”
“Credit card
change”
“App usability”
“Webview”
“Navigating
experience”
“Brightness
adjustment”

Source: ChatGPT supported/developed table.

Naeem et al. 7

Table 3. Third Step – ChatGPT Prompt, and Codes Selected by Humans and ChatGPT
Codes manually
ChatGPT-suggested codes with related selected by Naeem et
ChatGPT prompt to perform third step Justification of prompt keywords al. (2024a)

We now move to the third step: Coding. The aim Saldaña (2013) and Naeem et al. (2023) User Interface Satisfaction: “user- 1. Apathy
of coding is to use one or more words to jointly proposed definitions of coding. Codes should friendly”, “intuitive”, “organized” 2. Excitement and risks
represent a group of keywords and quotations be developed to present the combined Technical Frustrations: “glitchy”, “app 3. Gratification and
to answer the research question. Please take an meanings of related quotations and selected crashes”, “technical issues” capricious
abductive approach to developing potential keywords (Naeem et al., 2023). Budget-Conscious Shopping: “budget- 4. Utility and
meaningful codes, while considering the Codes offer new insights to answer the research friendly”, “savings”, “digital coupons” hinderance
following information and data set. These codes questions (Saldaña, 2013). Accessibility and Inclusion: “seniors”, 5. Absorption
should reflect the essence of participants’ points Codes can be done on base of inductive, “accessibility issues”, “friendly 6. Abandonment
of view in a conceptual form. Codes should deductive or abductive reasoning (Attride- associate”
consist of a maximum of three words. Stirling, 2001; Braun & Clarke, 2006; Creswell Customer Service Appreciation:
Please ensure that the suggested codes: & Poth, 2018; Elo & Kyngäs, 2008; Tavory & “customer service”, “service
• Contribute to answering the research Timmermans, 2014; Thomas, 2006). appreciation”, “customer feedback”
question and are developed through Coding should create robust, reflective, App Functional Reliability: “app
consideration of keywords and all selected resplendent, relevant, radical and righteous stability”, “app crashes”, “technical
quotations. codes, known as the 6 Rs of coding (Naeem et support”
• Reflect the 6 Rs coding framework – robust al., 2023). Convenience of Shopping:
(comprehensive to enable interpretation of the “convenience”, “scan and go”, “easy
data), reflective (reflect the selected keywords process”
and related quotations), resplendent (rich and Feedback and Suggestion: “feedback”,
conceptually powerful), relevant (related to the “user suggestion”, “app feedback”
research), radical (new contribution), and Membership and Loyalty: “membership
righteous (ethical and contextually correct). benefits”, “loyalty program”,
• Accommodate abductive reasoning by “membership verification”
integrating data-driven insights and theoretical Inventory Management Concerns:
frameworks. “stock levels”, “item availability”,
Data set segments: [Paste data set segments or “inventory accuracy”
specific excerpts] Shopping Experience Quality: “shopping
Identified keywords: [List identified keywords experience”, “customer experience
from Step 2] enhancement”, “in-store efficiency”
Selected quotations: [List identified quotations Navigation and Usability Issues:
from Step 1] “navigation problems”, “usability
Research question and aim: [Insert research issues”, “app navigation”
question and research aim]- (ChatGPT Checkout and Payment Process:
supported/developed prompt) “checkout process”, “digital
coupons”, “payment method”
Digital Integration and Updates: “app
updates”, “digital integration”, “app
improvement”
Environmental and Contextual
Adaptability: “environmental
adaptability”, “contextual
adaptability”, “store environment”
Engagement and Usage Barriers:
“engagement barriers”, “usage
barriers”, “barriers to use”

Source: ChatGPT supported/developed table.

understanding, and appreciation of the research approach is Once the AI is familiar with the research context and data,
important when selecting quotations (Eldh et al., 2020). the next step is to upload all of the transcripts and identify the
Oliver et al., 2005; Duranti, 2006). This is why ChatGPT research aim, objectives and research questions. It is also
was prompted/trained on the selected theories of the necessary to briefly provide the theoretical and philosophical
research, the research gap, research questions and aim of underpinning of the research. Importantly, the researcher
the research. Additionally, ChatGPT was provided with must ensure that the AI is fully trained on the research context
background information on the research, including the and systematic thematic analytical process which must be
number of selected organizations, the type of industry, and performed. The researcher should ask the AI to summarize the
the nature of the data. Table 1 outlines the ChatGPT systematic thematic analytical process introduced by Naeem
prompts used to acclimate it to the research, along with et al. (2023) before providing the prompts presented in
ChatGPT’s responses, including the quotes selected. Table 1. Once the researcher provides this information to the
8 International Journal of Qualitative Methods

Table 4. Fourth Step – ChatGPT Prompt, and Themes Selected by Human and ChatGPT

Themes manually
ChatGPT-generated themes based selected by Naeem
ChatGPT prompt to perform fourth step Rationale of prompt on selection of codes et al. (2024a)

Theming is a process of clustering codes Theming is the act of clustering 1. Digital Shopping Experience 1. Expectation and
based on their relationship to one codes based on their relatedness Enhancement perception.
another, in that they either fully or to each other, to answer research Relevant codes: user interface 2. Experience and
partially answer the research question. question(s ), holistically or satisfaction, difficult app for existence.
Given a set of codes and keywords, please partially (Naeem et al., 2023; shopping, shopping experience 3. Affordance
suggest appropriate themes to answer 2024b). quality, checkout and payment dichotomy.
the following research question and Themes need to be based within the process, digital integration and
achieve research aim. Your theme theoretical and philosophical updates.
development should be guided by the foundation of the research and 2. Operational and Technical
4 Rs framework of theming Reciprocal must closely link to the context of Efficiency
(connect the codes together in a the research to develop answers Relevant codes: technical
meaningful unique), Recognizable to the research questions frustrations, app functional
(reflect the actual data), Responsive (Bulmer, 1979; Coffey & reliability, inventory management
(responsive to the research question Atkinson, 1996; DeSantis & concerns, navigation and usability
and aim), Resourceful (rich to tell the Ugarriza, 2000). issues.
story of the research). Consideration of the 4 Rs of 3. Customer Engagement and
Make sure the themes are based on the theming: Reciprocal, Loyalty
following keywords and codes, and they Recognizable, Responsive and Relevant codes: customer service
helpfully answer the research question Resourceful (Naeem et al., 2023). appreciation, feedback and
and aim. suggestion, membership and
The proposed themes should be different loyalty.
from each other yet interconnected to 4. Accessibility and Inclusive
answer the research aim. Shopping
All suggested themes together need to Relevant codes: accessibility and
reflect the data appropriately, in terms inclusion, environmental and
of theoretical and conceptual adequacy, contextual adaptability,
and comprehensiveness in achieving the engagement and usage barriers.
research aim.
Use this information to create themes in
your mind.
• Coded data segments: [Insert the coded
data segments or any other collected
data]
• Identified keywords and codes: [paste
the list of identified keywords and codes
from the previous steps here].
• Research objective and question: [paste
the research objective and questions
here]>
• Theoretical underpinning: [paste the
theoretical underpinning here]
(ChatGPT aided /developed prompt)
(ChatGPT supported/developed prompt)
Source: ChatGPT supported/developed table.
Naeem et al. 9

Figure 3. Confirmation of AI Familiarization With Systematic Thematic Analysis Framework (Source: ChatGPT (AI) Response).
10 International Journal of Qualitative Methods

Table 5. Fifth Step – ChatGPT Prompt, and Concepts Selected by Humans and ChatGPT

ChatGPT- Concepts manually selected by

ChatGPT prompt to perform fifth step Rationale of prompt developed concepts Naeem et al. (2024a)

In the conceptualization step we aim to Development of concepts in relation • Digital Shopping The conceptual framework by Naeem
develop concepts based on our to the data and theoretical Experience et al. (2024a) outlined the
interpretation of themes, keywords underpinning (Babbie, 2016; Enhancement Affordances Flow Funnel, which
and codes. The concepts should link Byrne, 2015; Kerlinger & Lee, • Operational and has the following stages: Perceived
with theory in the context of this 2000; Naeem et al., 2023, 2024b). Technical Affordances, Actualization
research. Concepts should help Efficiency Affordances, Affordance
answer the research questions. Please • Customer Dichotomy
define each concept. The definitions Engagement and
should adhere to the principles of Loyalty
clarity, and cohere with the • Accessibility and
theoretical and philosophical context Inclusive Shopping
of this research. • Seamless
To guide this process, please consider Interaction
the following. • Ecosystem
• Research question: [Insert the • Engaged
specific research question] • Shopping
• Theoretical underpinnings: [Detail Community
the major factors of the theory] • Inclusive Shopping
• Themes: [Insert themes from • Adaptive
previous step] Affordance
Identified keywords and codes: [List the • User experience
identified keywords and codes from
previous steps] (ChatGPT supported/
developed prompt)
Source: ChatGPT supported/developed table.

AI, it is necessary to ensure that the AI understands the et al., 2024b), which has several advantages, including ex-
research context, and the systematic thematic analysis pro- panding current theory (Flick, 2014; Saldaña, 2013) and using
cess. In case of any discrepancies or inappropriate responses, participants’ own words to keep the analysis grounded in real
researchers can upload the following Figure 3 as an example data (Corbin & Strauss, 2008, 2015; Creswell & Poth, 2018;
of the familiarization stage for the AI. Morse, 2015; Richards, 2015; Ryan & Bernard, 2000). Table 2
shows the ChatGPT prompt used to ask the AI to select
keywords for systematic thematic analysis, along with the
Step 2: Selection of Keywords justification for the prompt, the keywords generated by
The second step of systematic thematic analysis is the selection ChatGPT, and the keywords manually selected by Naeem
of keywords, that is, the selection of meaningful words or et al. (2024a).
phrases from the transcript that capture participants’ rich ex- Table 2 shows that the large number of keywords identified
periences and insights to answer the research question (Naeem by the AI reflects its capacity to process large datasets
et al., 2023, 2024b). Keywords encapsulate the central and comprehensively, enabling it to identify a broader and more
meaningful ideas expressed by participants (Naeem et al., 2023, diverse range of relevant keywords. Consequently, different
2024b). The 6 Rs framework—Realness (reflecting the genuine words relating to the same issue, or the same words on dif-
experience of participants), Richness (meaningful, strong, ferent issues can capture a broader range of expressions and
powerful, and relevant to the research question), Repetition nuances than manually selected keywords typically allow.
(same word repeated or different words with the same meaning This AI capability, compared to human capabilities, empha-
used by participants), Rationale (strong reason or logic), sizes how the breadth of AI can complement the depth of
Repartee (linked to the context of the study), and Regal (crucial manually selected keywords. Table 2 presents a comparative
or important to understanding)—guides researchers in selecting analysis of keywords selected by ChatGPT, and those man-
impactful keywords. (Naeem et al., 2023). ually selected in Naeem et al.’s (2024a) study. It is evident that
The selection of keywords plays a vital role in the de- the AI’s keyword selection encompasses a greater number of
velopment of codes and uses the full potential of the transcript keywords, and a more comprehensive reflection of participant
rather than only the selected quotations (Naeem et al., 2024b). views, capturing both functional, customer experience and
In addition, keywords are used to interpret the data (Naeem experiential aspects of using shopping apps. For instance,
Naeem et al. 11

Table 6. Sixth Step – ChatGPT Prompt and Justification of Prompt

ChatGPT prompt to perform sixth step Justification of prompt

To the last step in the structured thematic analysis process described The conceptual model should be theoretically and philosophically
by Naeem et al. (2023), your goal is to synthesize the research grounded and evidenced based on actual data (Naeem et al., 2023).
findings into a conceptual model. Link these concepts with a figure These insights derived from a conceptual model should lead to both
or diagram that proposes a feasible solution to the following theoretical and practical understanding of the research problem,
research problem we want to address. The conceptual model must which, in turn, help guide decisions (Corley & Gioia, 2011; Smith
also contribute to managerial practices and following theory. The et al., 2009; Whetten, 1989; Whetten 1989, 1989).
proposed conceptual model should answer the following questions A good conceptual model should explain the how, why and what
additionally you need to keep in mind what are the key concepts of questions (Wertz et al., 2011)
exiting theory and what are the new concept developed through
this research? How are the newly developed concepts and the
already existing theory? What makes these concepts closely related
in the context of this research?
As you formulate the conceptual framework, consider the following.
• Research question and objectives: [Insert the specific research
question and objectives of the research]
• Theoretical underpinnings: [Detail the theoretical frameworks or
underpinnings that guide the study]
• Identified keywords, codes and themes: [List the keywords, codes,
and themes identified in previous steps]
• Philosophical underpinnings: [Include an explanation of the
epistemology –you can copy and paste the methodology section of
your research]
• Theoretical framework:Theoretical Framework: [Explain
theoretical frameworks used to guide the research]
The conceptual model must build a story that explains why the findings
are meaningful and how they work together to answer the research
question. *This content is tailored from OpenAI (ChatGPT
supported/developed prompt)

Source: ChatGPT supported/developed table.

ChatGPT includes terms like “Technical issues,” “Digital thematic analysis that can uncover the various usability
coupons,” Technical issues” “ExpressPay” “Location detec- aspects of the apps, potentially leading to richer insights.
tion” “Pickup” “Navigation problems” “Screen brightness” For example, the AI-based keywords, such as “Digital
and “Search functionality” which highlight specific features to coupons” and “Savings,” touch on the societal implications
understand the research issues in more detail. Therefore, the of using scan-and-go apps, providing insights into the
richness of the keywords, in terms of addressing different economic behaviors linked to the technological issues of
issues and using varied words for the same technological the apps. Therefore, researchers can compare both the
problem, can enhance the rigor of the research compared to manually selected keywords and the AI-generated key-
manually selected keywords. For example, in contrast to the words to repeat the different rounds in order to explore the
manually selected keywords related to technological issues, problem from a range of angles. For example, in the case of
such as “Garbage app,” “not working,” “looking to increase the selected case study, technological, social, and shopping
buying,” “data could be lost,” “information/could not find,” experience angles can be further explored through findings
and “doesn’t sync in real time,” which are less rigorous, the that include richer words or phrases aligned with these
keywords selected by ChatGPT offer a broader and more angles, based on the 6 Rs (Naeem et al., 2023). This ensures
comprehensive range. the robustness and relevance of the keywords related to the
While there is some diversity in the manually selected research problem. Consequently, this capability highlights
keywords to understand other perspectives, such as “not my AI’s potential to strengthen the traditional keyword se-
job,” “job-eating app,” and “extra burden,” these primarily lection process. Such extensive keyword identification
focus on the subjective drawbacks perceived and experi- introduces new dimensions at the coding stage, enhancing
enced by users, reflecting their reluctance to use the apps. In the richness of codes and allowing for the exploration of the
contrast, the broader selection of keywords by the AI, problem through different angles. This will be covered in
which includes terms like “Reliability issues” and “App the next section, which addresses how a larger number of
review request,” allows for a more meaningful and rich keywords provides more opportunities to strengthen
12 International Journal of Qualitative Methods

existing codes and generate new ones, ultimately improving keywords provide valuable insights into the specific technical
the richness of the study. challenges users encounter during their shopping experiences.
These challenges are associated with the frustration users may
feel using app, and the subsequent abandonment of the app.
Step 3: Coding
The AI also generated more customer shopping experience–
Coding is the systematic process (Saldaña, 2013) of labelling related codes, such as Shopping Convenience, In-store Pickup,
related data different segments with short phrases or words that Warehouse Deals, and Inventory Management Concerns,
lend conceptual meaning to the data (Naeem et al., 2023). which highlight operational efficiencies within the organiza-
Boyatzis (1998, p., 63) state that a code is the “most basic tion, including stock levels and product availability. . These
segment, or element, of the raw data or information that can be advanced codes make it possible to study the challenges
assessed in a meaningful way regarding the phenomenon”. customers experience which are more related to organization
Codes should be developed to present the combined meanings support and infrastructure. The above analysis of the depth of
of related quotations and selected keywords; they act as the keywords used to develop 16 codes provides sufficient evi-
conceptual backbone of the analysis (Naeem et al., 2023). dence to suggest that AI can add value in terms of the ro-
Codes unearth patterns and relationships between keywords bustness and richness of coding. AI can enhance the richness
(Naeem et al., 2023). They enable flexibility and evolution in of codes either by supporting them with additional keywords
research in response to new conceptual insights to answer or by generating further codes that address other aspects of the
research questions (Saldaña, 2013). Inductive, deductive or research problem.
abductive reasoning can be used to develop codes. Deductive
reasoning makes an inference based on facts or theories. In-
Step 4: Theme Development
ductive reasoning makes an inference based on the observation
of patterns emerging from the data. Abductive reasoning Theme development is about clustering codes into different
generates a hypothesis to explain observations about emerging categories according to their relevance in the context of the
patterns that align with theory (Attride-Stirling, 2001; Braun & research question. Consequently, theming is about structuring
Clarke, 2006; Creswell & Poth, 2018; Elo & Kyngäs, 2008; the data to form a meaningful conceptual narrative to achieve
Tavory & Timmermans, 2014; Thomas, 2006). the research aim (Naeem et al., 2023, 2024b). Yet these themes
The coding process places emphasis on creating robust, need to be grounded, with real data, participant language
reflective, resplendent, relevant, radical and righteous codes, (Naeem et al., 2023), and the theoretical and philosophical
which is known as the 6 Rs of coding (Naeem et al., 2023). underpinnings of the research (Coffey & Atkinson, 1996;
Moreover, Naeem et al. (2023) stated that researchers need to DeSantis & Ugarriza, 2000). Thus, themes are a way of de-
consider the grouping of keywords to make them more riving meaning from tied codes (Charmaz, 2006; DeSantis &
meaningful in the analysis. Table 3 shows the coding Ugarriza, 2000) to develop a conceptual framework. Naeem
guidelines given to ChatGPT along with a justification of the et al., (2023) continued to propose that themes need to follow
guidelines. It also identifies the codes generated by ChatGPT the 4 Rs of theming. They must be: Reciprocal (joining the
and the codes manually selected by Naeem et al. (2024a). codes in meaningful form); Recognizable (from the real data
ChatGPT suggests 16 codes, whereas Naeem et al. (2024a) so they can be identifiable); Responsive (to the research
only developed six codes. This outcome shows that ChatGPT question, as well as its aim); and Resourceful (rich enough to
can enhance the use of all potential data. Their can therefore be be coherent as a research story). This ensures a multidi-
argued that manual codes are more theoretical than ChatGPT mensional connection between real data, the research aim and
codes. Indeed, ChatGPT was instructed to take an abductive objectives, the methodological underpinning and the theo-
approach, which led to the codes being based on common retical underpinnings of research (Naeem et al., 2023). Based
patterns in the data. ChatGPT can be instructed to take either on these guidelines, a ChatGPT prompt (see Table 4) was used
an inductive or deductive approach to the development of to ask ChatGPT to generate potential themes.
codes. The ChatGPT prompts to identify codes from the data The following prompt was developed and generated by
are shown in Table 3 below. ChatGPT, based on the perspectives of various authors. This
Table 3 provides a comparison between manual coding and prompt was used to ask ChatGPT to suggest theme names
AI-generated coding. As evident, Naeem et al. (2024a) based on the keywords and codes identified in the earlier
identified 6 codes, while AI-generated coding produced 16 stages of analysis. The themes generated by ChatGPT are
codes from the data. Therefore, two advantages emerge for presented in Table 4 and compared with both the AI-generated
researchers, as the AI-developed codes are much richer than themes and those manually selected by Naeem et al. (2024a).
the manually generated ones.The second benefit of more and As above table illustrates, the implementation of AI-
richer keywords selected by the AI means that more codes (16 developed themes has largely broadened and deepened the
in this case) can be developed providing a rich understanding themes themselves as compared to those that were developed
of the research problem. AI codes like “Technical Frustra- manually. This also illustrates how the theme of Customer
tions” and “App Functional Reliability” with support the Engagement and Loyalty, as discussed throughout the AI
Naeem et al. 13

analysis, aligns with the Affordance Dichotomy theme iden- together and answer the research question. A ChatGPT prompt
tified by Naeem et al. (2024a), contributing to the existing (see Table 5) based on these guidelines was developed to ask
body of knowledge on gaining value from customers through ChatGPT to suggest concepts.
appreciation and obtaining feedback that organisations can Table 5 shows the differences or similarities between AI-
leverage further. The AI-generated themes such as “Acces- generated, and manually developed concepts. AI based no-
sibility and Inclusive Shopping.” do not explicitly appear in tions like “Digital Shopping Experience,” “Operational and
Naeem et al., (2024a). Thus, the themes derived manually did Technical Efficiency” and “Customer Engagement and Loy-
not account for environmental adaptability and contextual alty” formulated much wider and comprehensive explanation
impediments, and as a result, these aspects were not identified. than the manual conceptualization that was developed man-
. Furthermore, manual themes excluded certain technological ually by Naeem et al. (2024a). Additionally, AI-generated
features. AI suggested precise and granular themes, such as ideas, like “Ecosystem” and “Engaged Shopping Commu-
Digital Shopping Experience and Operational Technical Ef- nity,” might add more theoretical concepts to the notion of
ficiency, which reflect action-oriented and system-focused affordance, extending some communal and environmental
features, highlighting specific technology-related user per- aspects of the shopping experience. Each of these concepts
spectives. Therefore, the diverse range of AI-generated codes generated by the AI is seemingly a rich amalgamation of the
should be considered to enhance the breadth of manual codes codes and themes already discovered, showing that the codes
and may contribute to the development of more significant and themes are embedded within the data, as well as the
themes that enrich the study. methodological, philosophical and theoretical foundations of
the research. It is evident that the ideas generated by AI are
rich in detail and offer deeper analysis of the affordances of
Step 5: Conceptualization
apps, particularly in relation to seamless user experience and
The fifth step, conceptualization, involves creating a link operational efficiency. These could expand the boundaries of
between theory and the developed themes through inter- the conceptual framework. This is an indication that though AI
preting keywords and codes for conceptual clarification and manual analyses are not the same, a combination of both
(Naeem et al., 2023). Concepts are generalized ideas that represents a better result. However, if the researcher instructs
represent specific phenomena or objects (Kerlinger & Lee, AI to align the main themes and codes with existing theories
2000). Conceptualization plays a crucial role in theoretical and models, this can be achieved by guiding ChatGPT to
development by categorizing data into meaningful forms develop concepts in line with those frameworks. Thus, the use
through linking themes with the context of the research of AI can significantly enhance the conceptual foundation,
(Naeem et al., 2023). The interpretation of themes can be both by correlating with established models and by intro-
authenticated by using participants’ language, which en- ducing new aspects and finer levels of nuance.
ables systematic investigation (Babbie, 2016; Byrne, 2015;
Naeem et al., 2023, 2024a). In the case of deductive rea-
soning, the research needs to focus on concepts that are
Step 6: Development of the Conceptual Framework
underpinned by theory, while inductive reasoning can re- The last stage of systematic thematic analysis is to produce a
inforce the concepts through the use of the participants’ conceptual model/framework that brings together the identi-
language (Babbie, 2016; Byrne, 2015). fied concepts with each other. These links should connect with
Naeem et al. (2023) provided guidelines in the form of the data as well as with the existing literature and theoretical
questions to evaluate conceptual definitions: Are the defini- basis for the research (Naeem et al., 2023) to offer a solution to
tions clear and explicit in a theoretical context? Do the new the selected problem. Therefore, a conceptual model should be
concepts enhance understanding of the research findings? Are constructed by combining the thematic synthesis of codes and
they theoretically and philosophically accurate, and are they keywords alongside theoretical and philosophical discussions
traceable from the primary data? Do they reflect the actual of the research (Naeem et al., 2023). The conceptual model/
data? Are the concepts grounded philosophically, analytically framework justifies findings through the development and
and theoretically? Are they appropriate to justify the research linkage of new concepts (Lewis, 1998). This includes sto-
outcomes in relation to theory and practice? Are the concepts rytelling and process (Naeem et al., 2023), providing a fig-
theoretically interconnected with the development of a con- urative, conceptual, and narrative description of the properties
ceptual model? and dimensions of the identified codes and themes, as they
These guidelines were used to develop a ChatGPT prompt relate to a selected theoretical or practical solution for the
to ask for AI assistance with the conceptualization of themes. research problem. Therefore, the conceptual framework must
Table 5 shows that the concepts developed by ChatGPT are offer clear and intuitive guidance for practitioners, assisting
based on the participants’ experiences, whereas the manually them in complex, applied decision-making (Corley & Gioia,
derived concepts are more related to the theoretical under- 2011; Hamilton et al., 2023; Smith et al., 2009; Whetten,
pinnings. Researchers could amend the prompt and ask 1989). Thus, the framework should incorporate a process that
ChatGPT to develop theory-related concepts that link themes professionals can follow to transfer research into practice. The
14 International Journal of Qualitative Methods

concepts in the conceptual model need to connect with ex- criticised for simplifying complex issues (Vaismoradi et al.,
isting theoretical concepts and the study’s theoretical rationale 2016) and lacking methodological consistency throughout
should be noted (Reay & Whetten, 2011; Wertz et al., 2011) to the analytical process (Nowell et al., 2017; Vaismoradi et al.,
connect newly developed concepts with existing theoretical 2016). AI-powered thematic analysis of large amounts of
concepts (Naeem et al., 2023; Rynes, 2002). While in general, data could be less time consuming, and AI is able to analyse
the merit of theory is measured by its contribution to complex patterns in data (Christou, 2023b; Lee & Choi,
knowledge (Chun Tie et al., 2019; Thurlow, 2020), “The route 2023). Many researchers have indicated that AI use in
to good theory leads not through gaps in the literature but research is more time-efficient and reduces human bias. It
through an engagement with problems in the world” (Kilduff, also improves the efficiency and accuracy of the
2006, p. 252). Moreover, the principle of good theory involves analysis (Christou, 2023; Lee & Choi, 2023).
identifying what relates to what (What), understanding the The challenges in generating AI instructions, the necessity
interconnectivity between these concepts (How), explaining for verification of results, and acknowledgement of the bias
why (linking rationales) (Why), and finding the balance be- inherent in AI-generated data underscore the need for rigorous
tween comprehensiveness and parsimony when solving a oversight of AI usage (Zhang et al., 2023). As stated above,
specific problem (Dubin, 1978). Therefore, Naeem et al. some of the challenges traditional thematic analysis faces can
(2023) suggest that the conceptual framework should de- be overcome by using AI, but AI-powered thematic analysis is
fine and address the selected research problem. A prompt not without its challenges or weaknesses. For example, AI can
for ChatGPT was developed based on the viewpoints of the reduce human bias in analysis, but AI algorithmic bias can
above authors, requesting ChatGPT to provide explanations undermine equity and validity (Noble, 2020), which could
of the concepts to be connected in a diagram format, along have an impact on research findings. The lack of visibility of
with advice on a conceptual model that would explain these AI inputs and processes, which is termed the “black box”
concepts. The proposed conceptual model, based on in- problem, creates challenges of transparency and account-
structions given to ChatGPT, and ChatGPT’s suggested ability (Greene et al., 2021) that require human intervention.
conceptual model, can be adjusted or adapted by re- In response, this paper proposes a prompt for each step of
searchers. Table 6 shows the ChatGPT prompt used to systematic thematic analysis that could reduce human bias
request the AI to establish a conceptual model that syn- while, at the same time, improving transparency and ac-
thesizes the findings from the research, along with the countability. Indeed, accountability is provided by the ratio-
justification for the prompt. nale for each of the prompts during the analysis process.
Another issue with using AI in research is researcher influ-
Integrating AI in Thematic Analysis: Dealing ence; that is, the instructions given to an AI system by a
researcher (Christou, 2024), and lack of reliability and ac-
with Bias and Navigating Challenges curacy (Christou, 2023b). Some traditional methods can be
This paper describes a toolkit (ChatGPT prompts) for sys- applied to deal with the consistency and accuracy of codes and
tematic thematic analysis (Naeem et al., 2023) that uses AI. themes, such as member-checking (Creswell, 2014; Ngulube,
The toolkit helps researchers thematically analyse data on the 2015) and repeating multiple coding rounds (Sweeney et al.,
basis of various methodological approaches. In general, 2013). In addition, this paper provides guidelines on providing
thematic analysis is limited by aspects such as subjectivity and the rationale behind the instructions given to AI, and sufficient
potential biases due to the limited ability of researchers to information should be provided to the AI system at the first
process large amounts of data (Morgan, 2023). However, this step of systematic thematic analysis (familiarization) to inform
paper introduces a toolkit that can consider large datasets, it about the research context and data.
varied research contexts, and complex theoretical, methodo- The toolkit described in this paper not only benefits systematic
logical and philosophical underpinnings at each stage of thematic analysis but also enriches other qualitative research
systematic thematic analysis (Christou, 2023a). In addition, methods. For instance, the initial coding in narrative and content
AI-powered systematic processes could improve the consis- analysis in ethnographic research, which focuses on the identi-
tency of thematic analysis. Turobov et al., (2024) indicated fication of patterns in behaviour, can be achieved through the
that, traditionally, thematic analysis has been criticised for its application of the coding guidelines outlined in this paper.
inconsistency and lack of generalizability. Phenomenology focuses on the meanings individuals attach to
Traditional thematic analysis faces other issues that could their experience (Moustakas, 1994), anthropology requires un-
also be overcome by the use of AI-powered systematic derstanding of the cultural setting (Geertz, 1973) and grounded
thematic analysis. For example, researchers have highlighted theory builds theory (Charmaz, 2014; Glaser & Strauss, 1999).
that traditional thematic analysis variously: lacks a theo- The toolkit espoused here can be used to identify patterns in data
retical underpinning (Hamilton et al., 2023), is subject to that could be helpful to categorize data into themes or codes, and
research biases (Ngulube, 2015; Pope et al., 2000), is dif- participants’ language could be used to develop codes and
ficult to use when analysing large amounts of data, and is keywords that will facilitate interpretation of the results in many
time consuming (Nowell et al., 2017). It has also been types of qualitative research.
Naeem et al. 15

Conclusion, Contribution and qualitative research methods by exploring research ques-

Recommendations for Future Research tions such as:

The current paper introduces an AI toolkit which draws on 1. How can AI be used to ensure data saturation in
previous research. It provides a rigorously developed AI toolkit systematic thematic analysis to extract more insights
to enhance transparency around the use of AI in research. from source data?
Qualitative researchers can develop their own toolkits by 2. How can AI be used to improve the authenticity, rigour,
considering the process set out in this paper. In addition, the reliability, trustworthiness, transparency and account-
paper provides guidelines on how to develop a toolkit for AI to ability of qualitative research in thematic analysis?
reduce bias and improve the reliability of qualitative research. 3. How can the process of reflexivity be applied to reflect
When examining the future of qualitative data analysis, two on AI-based systematic thematic analysis? What are
critical truths are becoming more apparent: The first is that AI the major factors and steps that should be considered
must play an integral role in the process of improving the when using AI in systematic thematic analysis to
discoveries of qualitive research. The second is that although AI improve the transparency of research?
is a powerful tool that can boost and redress many aspects of 4. How can AI be used to develop case study as foundation
qualitative data analysis, it makes us more aware of the different of systematic thematic analysis in the qualitative research?
ways in which problems can originate in our desire to produce 5. How can AI-based systematic thematic analysis sup-
enhanced, ethical and insightful results. port ethnography, phenomenology, anthropology,
This paper represents a pioneering attempt to integrate an grounded theory, and action and narrative research?
AI, specifically ChatGPT, into the systematic thematic 6. How can researchers enhance cognitive input into
analysis process introduced by Naeem et al. (2023). This generative AI to improve the quality of analysis in
toolkit entwines AI with each of the six steps of thematic qualitative research?
analysis, adding to the richness, robustness and efficiency of
thematic analysis. Importantly, the typologies of ChatGPT
Acknowledgements
prompts developed though consideration of different authors
can enhance the performance of AI-powered analysis with The authors would like to acknowledge the use of ChatGPT in the
implications for the quality of research. AI can analyse development of this manuscript. ChatGPT was utilized to generate
extensive sets of data, helping recognise patterns, keywords, prompts for each stage of systematic thematic analysis, developed
and themes that would otherwise be missed or take an in- based on the authors’ differing viewpoints and to prompt the six
ordinate amount of time to uncover. As such, AI adds a new steps of systematic thematic analysis introduced by Naeem et al.
depth of analysis to large data sets through consideration of (2023). This analysis was then compared with the case study se-
the methodological and theoretical underpinning of the study. lected from Naeem et al. (2024a). Additionally, ChatGPT aided in
This supports the idea that AI can supplement and enhance enhancing the clarity of language and in the structuring and re-
traditional thematic analysis processes and could be helpful to structuring of sentences throughout the manuscript. The authors
introduce a theory or conceptual model that is richer, and far more have rigorously reviewed and revised the entire manuscript, as-
nuanced than traditional/ human based thematic analysis. suming full responsibility for the accuracy of the content presented.
However, this does not mean that AI alone can perform the Furthermore, the manuscript underwent professional proofreading,
thematic analysis. The researcher needs to be fully involved in the and a certificate of proofreading was provided to the editor during
process, repeating any stages of the analysis and providing ad- the review process.
ditional details to AI to achieve the research aim. AI should be
trained about the research and the thematic analysis process, as Statements and Declarations
demonstrated in Figures 1 and 2 above. Once the researcher
receives consistent responses and AI becomes familiar with the Funding
The authors received no financial support for the research, authorship,
research and systematic thematic analysis, the researcher can
and/or publication of this article.
begin to perform the analysis.
For the sake of transparency, the authors would like to Conflicting Interests
acknowledge the use of ChatGPT was employed to enhance The authors declared no potential conflicts of interest with respect to
clarity of language, improving understanding and sentence the research, authorship, and/or publication of this article.
structure. Nonetheless, the authors have gone through the
entire manuscript and take absolute responsibility for the References
information presented. The paper also provides recom- Attride-Stirling, J. (2001). Thematic networks: An analytic tool for
mendations on how researchers can mitigate these limita- qualitative research. Qualitative Research, 1(3), 385–405.
tions using different traditional strategies. The scope of this https://doi.org/10.1177/146879410100100307
paper is limited to the use of AI for systematic thematic Babbie, E. R. (2016). The practice of social research (14th ed.).
analysis. Future researchers could extend the scope to other Cengage Learning.
16 International Journal of Qualitative Methods

Boyatzis, R. E. (1998). Transforming qualitative information: Currie, G., Singh, C., Nelson, T., Nabasenja, C., Al-Hayek, Y., & Spuur, K.
Thematic analysis and code development. Sage. (2023b). ChatGPT in medical imaging higher education. Radiog-
Braun, V., & Clarke, V. (2006). Using thematic analysis in psy- raphy, 29(4), 792–799. https://doi.org/10.1016/j.radi.2023.05.011
chology. Qualitative Research in Psychology, 3(2), 77–101. Davidson, C. (2018). Reflection/commentary on a past article:
https://doi.org/10.1191/1478088706qp063oa “Transcription: Imperatives for qualitative research”. Interna-
Braun, V., & Clarke, V. (2022). Toward good practice in thematic tional Journal of Qualitative Methods, 17(1), 160940691878822.
analysis: Avoiding common problems and be(com)ing a knowing https://doi.org/10.1177/160940690900800206
researcher. International Journal of Transgender Health, 24(1), De Paoli, S. (2024a). Performing an inductive thematic analysis of
1–6. https://doi.org/10.1080/26895269.2022.2129597 semi-structured interviews with a large language model: An
Bulmer, M. (1979). Concepts in the analysis of qualitative data. The exploration and provocation on the limits of the approach.
Sociological Review, 27(4), 651–677. https://doi.org/10.1111/j. Social Science Computer Review, 42(4), 997–1019. https://doi.
1467-954X.1979.tb00354.x org/10.1177/08944393231220483
Byrne, D. (2015). Response to fugard and potts: Supporting De Paoli, S. (2024b). Further explorations on the use of large lan-
thinking on sample sizes for thematic analyses: A quanti- guage models for thematic analysis. Open-ended prompts, better
tative tool. International Journal of Social Research Meth- terminologies and thematic maps. Forum Qualitative So-
odology, 18(6), 689–691. https://doi.org/10.1080/13645579. zialforschung/Forum: Qualitative Social Research, 25(3).
2015.1005455 https://doi.org/10.17169/fqs-25.3.4196
Charmaz, K. (2006). Constructing grounded theory: A practical DeSantis, L., & Ugarriza, D. (2000). The concept of theme as used in
guide through qualitative analysis. Sage. qualitative nursing research. Western Journal of Nursing Research,
Charmaz, K. (2014). Constructing grounded theory. Sage. 22(3), 351–372. https://doi.org/10.1177/019394590002200308
Christou, P. (2023a). A critical perspective over whether and how to Dubin, R. (1978). Theory building. Free Press.
acknowledge the use of artificial intelligence (AI) in qualitative Du Bois, J. W. (1991). Transcription design principles for spoken
studies. Qualitative Report, 28(7), 1981–1991. discourse research. Pragmatics. Quarterly Publication of the
Christou, P. (2023b). Ηow to use artificial intelligence (AI) as a International Pragmatics Association (IPrA), 1(1), 71–106.
resource, methodological and analysis tool in qualitative https://doi.org/10.1075/prag.1.1.04boi
research? Qualitative Report, 28(7), 1968–1980. https://doi.org/ Duranti, A. (2006). Transcripts, like shadows on a wall. Mind,
10.46743/2160-3715/2023.6407 Culture and Activity, 13(4), 301–310. https://doi.org/10.1207/
Christou, P. A. (2024). Thematic analysis through artificial intelli- s15327884mca1304_3
gence (AI). The Qualitative Report, 29(2), 560–576. https://doi. Eldh, A. C., Årestedt, L., & Berterö, C. (2020). Quotations in
org/10.46743/2160-3715/2024.7046 qualitative studies: Reflections on constituents, custom, and
Chun Tie, Y., Birks, M., & Francis, K. (2019). Grounded theory purpose. International Journal of Qualitative Methods, 19, 1.
research: A design framework for novice researchers. Sage https://doi.org/10.1177/1609406920969268
Open Medicine, 7(3), 2050312118822927. https://doi.org/10. Elo, S., & Kyngäs, H. (2008). The qualitative content analysis
1177/2050312118822927 process. Journal of Advanced Nursing, 62(1), 107–115. https://
Coffey, A., & Atkinson, P. (1996). Making sense of qualitative data: doi.org/10.1111/j.1365-2648.2007.04569.x
Complementary research strategies. Sage. Flick, U. (2014). An introduction to qualitative research. Sage.
Corbin, J. M., & Strauss, A. L. (2008). Basics of qualitative research: Geertz, C. (1973). Thick Description: Toward an Interpretive Theory
Techniques and procedures for developing grounded theory (3rd of Culture. In The Interpretation of Cultures: Selected Essays
ed.). Sage. (pp. 3–30). New York: Basic Book.
Corbin, J. M., & Strauss, A. L. (2015). Basics of qualitative research: Glaser, B., & Strauss, A. (1999). Discovery of grounded theory:
Techniques and procedures for developing grounded theory (4th Strategies for qualitative research (1st ed.). Routledge.
ed.). Sage. Glaser, B. G., & Strauss, A. L. (2009). The discovery of grounded
Corley, K. G., & Gioia, D. A. (2011). Building theory about theory theory: Strategies for qualitative research (4th ed.). Aldine.
building: What constitutes a theoretical contribution? Academy Greene, D., Hoffmann, A. L., & Stark, L. (2021). Better, nicer,
of Management Review, 36(1), 12–32. https://doi.org/10.5465/ clearer, fairer: A critical assessment of the movement for ethical
amr.2009.0486 artificial intelligence and machine learning. Proceedings of the
Creswell, J. W. (2014). Research design: Qualitative, quantitative, 2021 ACM Conference on Fairness, Accountability, and
and mixed methods approach. Sage. Transparency. https://scholarspace.manoa.hawaii.edu/server/
Creswell, J. W., & Poth, C. N. (2018). Qualitative inquiry and api/core/bitstreams/849782a6-06bf-4ce8-9144-a93de4455d1c/
research design: Choosing among five approaches (4th ed.). content (Accessed at 02-09-2024).
Sage Publications. Hamilton, L., Elliott, D., Quick, A., Smith, S., & Choplin, V. (2023).
Currie, G., Singh, C., Nelson, T., Nabasenja, C., Al-Hayek, Y., & Exploring the use of AI in qualitative analysis: A comparative
Spuur, K. (2023a). ChatGPT in medical imaging higher edu- study of guaranteed income data. International Journal of
cation. Radiography, 29(4), 792–799. https://doi.org/10.1016/j. Q u a l i t a t i v e M e t h o d s , 2 2 . h t t p s : / / d o i . o rg / 1 0 . 11 7 7 /
radi.2023.05.011 16094069231201504
Naeem et al. 17

Huang, J., & Tan, M. (2023). The role of ChatGPT in scientific Naeem, M., & Ozuem, W. (2025). Understanding social motivation
communication: Writing better scientific review articles. in scan and go technology. Journal of Computer Information
American Journal of Cancer Research, 13(4), 1148–1154. Systems, 65, 1–18. https://doi.org/10.1080/08874417.2024.
Jaffe, A. (2000). Introduction: Non-standard orthography and non- 2445007
standard speech. Journal of Sociolinguistics, 4(4), 497–513. Naeem, M., Ozuem, W., Howell, K., & Ranfagni, S. (2023). A step-
https://doi.org/10.1111/1467-9481.00127 by-step process of thematic analysis to develop a conceptual
Jaffe, A. (2007). Variability in transcription and the complexities of model in qualitative research. International Journal of Quali-
representation, authority, and voice. Discourse Studies, 9(6), tative Methods, 22(11), 1–18. https://doi.org/10.1177/
831–836. https://doi.org/10.1177/1461445607082584 16094069231205789
Kerlinger, F. N., & Lee, H. B. (2000). Foundations of behavioral Naeem, M., Ozuem, W., Howell, K., & Ranfagni, S. (2024a). The
research. Harcourt College Publishers. conceptualization of enablers and constraints of in-store buying
Kilduff, M. (2006). Editor’s comments: Publishing theory. Academy as part of the affordances flow funnel process through scan and
of Management Review, 31(2), 252–255. https://doi.org/10. go apps. Psychology & Marketing, 41(5), 1060–1081. https://
5465/amr.2006.20208678 doi.org/10.1002/mar.21968
Laios, A., Theophilou, G., De Jong, D., & Kalampokis, E. (2023). Naeem, M., Ozuem, W., Howell, K., & Ranfagni, S. (2024b). De-
The future of AI in ovarian cancer research: The large language mystification and actualisation of data saturation in qualitative
models perspective. Cancer Control: Journal of the Moffitt research through thematic analysis. International Journal of
Cancer Center, 30, 10732748231197915. https://doi.org/10. Qualitative Methods, 23(1), 1–17. https://doi.org/10.1177/
1177/10732748231197915 16094069241229777
Lee, S. W., & Choi, W. J. (2023). Utilizing ChatGPT in clinical Naeem, N., Ozuem, W., Ranfagni, S., & Howell, K. (2025). User
research related to anesthesiology: A comprehensive review of generated content and brand engagement: Exploring the role of
opportunities and limitations. Anesthesia and Pain Medicine, electronic semiotics and symbolic interactionism on Instagram.
18(3), 244–251. https://doi.org/10.17085/apm.23056 Computers in Human Behavior, 168, 108642. https://doi.org/10.
Lee, V. V., van der Lubbe, S. C., Goh, L. H., & Valderas, J. M. (2024). 1016/j.chb.2025.108642
Harnessing ChatGPT for thematic analysis: Are we ready? Ngulube, P. (2015). Qualitative data analysis and interpretation:
Journal of Medical Internet Research, 26, Article e54974. Systematic search for meaning. In E. R. Mathipa & M. T.
https://doi.org/10.2196/54974 Gumbo (Eds.), Addressing research challenges: Making
Lewis, M. W. (1998). Iterative triangulation: A theory development headway for developing researchers (pp. 131–156). Mosala-
process using existing case studies. Journal of Operations MASEDI Publishers & Booksellers.
Management, 16(4), 455–469. https://doi.org/10.1016/s0272- Noble, S. U. (2020). Algorithms of oppression: How search engines
6963(98)00024-2 reinforce racism. NYU Press.
Maxwell, J. A. (2013). Qualitative research design: An interactive Nowell, L. S., Norris, J. M., White, D. E., & Moules, N. J. (2017).
approach. Sage. Thematic analysis: Striving to meet the trustworthiness criteria.
Meyer, J. G., Urbanowicz, R. J., Martin, P. C. N., O’Connor, K., Li, International Journal of Qualitative Methods, 16(1), 1–13.
R., Peng, P.-C., Bright, T. J., Tatonetti, N., Won, K. J., Gonzalez- Ochs, E. (1979). Transcription as theory. In E. Ochs & B. B Schiefflin
Hernandez, G., & Moore, J. H. (2023). ChatGPT and large (Eds.), Developmental pragmatics (pp. 43–72). Academic.
language models in academia: Opportunities and challenges. Ochs, E. (1999). In A. Jaworski & N. Coupland (Eds.), The discourse
BioData Mining, 16(1), 20. https://doi.org/10.1186/s13040- reader (pp. 158–166). Routledge.
023-00339-9 Oliver, D. G., Serovich, J. M., & Mason, T. L. (2005). Constraints and
Mondada, L. O. R. E. N. Z. A. (2007). Commentary: Transcript opportunities with interview transcription: Towards reflection in
variations and the indexicality of transcribing practices. Dis- qualitative research. Social Forces; a Scientific Medium of
course Studies, 9(6), 809–821. https://doi.org/10.1177/ Social Study and Interpretation, 84(2), 1273–1289. https://doi.
1461445607082581 org/10.1353/sof.2006.0023
Morgan, D. L. (2023). Exploring the use of artificial intelligence for Pope, C., Ziebland, S., & Mays, N. (2000). Qualitative research in
qualitative data analysis: The case of ChatGPT. International health care: Analyzing qualitative data. BMJ, 320(7227),
Journal of Qualitative Methods, 22(6). https://doi.org/10.1177/ 114–116. https://doi.org/10.1136/bmj.320.7227.114
16094069231211248 Prescott, M. R., Yeager, S., Ham, L., Rivera Saldana, C. D., Serrano,
Morse, J. M. (2015). Critical analysis of strategies for determining V., Narez, J., Paltin, D., Delgado, J., Moore, D. J., & Montoya, J.
rigor in qualitative inquiry. Qualitative Health Research, 25(9), (2024). Comparing the efficacy and efficiency of human and
1212–1222. https://doi.org/10.1177/1049732315588501 generative AI: Qualitative thematic analyses. JMIR AI, 3, Ar-
Moustakas, C. (1994). Phenomenological research methods. Sage. ticle e54482. https://doi.org/10.2196/54482
Naeem, M. (2025). Emerging trends in global E-retailing: Exploring Reay, T., & Whetten, D. A. (2011). What constitutes a theoretical
the dark side of scan and go in-store technologies in consumer contribution in family business? Family Business Review, 24(2),
shopping journeys. International Marketing Review. https://doi. 105–110. https://doi.org/10.1177/0894486511406427
org/10.1108/imr-06-2023-0110 Richards, L. (2015). Handling qualitative data: A practical guide. Sage.
18 International Journal of Qualitative Methods

Ryan, G., & Bernard, H. (2000). Data management and analysis Tavory, I., & Timmermans, S. (2014). Abductive analysis: Theorizing
methods. In N. Denzin & Y. Lincoln (Eds.), Handbook of qualitative research. University of Chicago Press.
qualitative research (2nd ed., pp. 769–802). Sage. Thomas, D. R. (2006). A general inductive approach for ana-
Rynes, S. (2002). From the editors: Some reflections on contribution. lyzing qualitative evaluation data. American Journal of
Academy of Management Journal, 45(2), 311–313. https://doi. Evaluation, 27(2), 237–246. https://doi.org/10.1177/
org/10.5465/amj.2002.17571225 1098214005283748
Saldaña, J. (2013). The coding manual for qualitative researchers. Sage. Thorne, S. (2000). Data analysis in qualitative research. Evidence
Salimi, A., & Saheb, H. (2023). Large language models in oph- Based Nursing, 3(3), 68–70. https://doi.org/10.1136/ebn.3.3.68
thalmology scientific writing: Ethical considerations blurred Thurlow, L. (2020). Grounded theory and the PhD – notes for
lines or not at all? American Journal of Ophthalmology, 254, novice researchers. Journal of Humanities and Applied So-
177–181. https://doi.org/10.1016/j.ajo.2023.06.004 cial Sciences, 2(4), 257–270. https://doi.org/10.1108/jhass-
Semrl, N., Feigl, S., Taumberger, N., Bracic, T., Fluhr, H., Blockeel, 05-2020-0079
C., & Kollmann, M. (2023). AI language models in human Turobov, A., Coyle, D., & Harding, V. (2024). Using ChatGPT for
reproduction research: Exploring ChatGPT’s potential to assist thematic analysis. Bennett Institute for Public Policy, University
academic writing. Human Reproduction (Oxford, England), of Cambridge. Working Paper [Preprint]. https://www.
38(12), 2281–2294. https://doi.org/10.1093/humrep/dead105 researchgate.net/publication/380607847_Using_ChatGPT_
Slembrouck, S. T. E. F. (2007). Transcription – the extended di- for_Thematic_Analysis (Accessed at 01-10-2024).
rections of data histories: A response to M. Bucholtz’s ‘variation Vaismoradi, M., Jones, J., Turunen, H., & Snelgrove, S. (2016).
in transcription’. Discourse Studies, 9(6), 822–827. https://doi. Theme development in qualitative content analysis and thematic
org/10.1177/1461445607082582 analysis. Journal of Nursing Education and Practice, 6(5),
Smith, J. A., Flowers, P., & Larkin, M. (2009). Interpretative phe- 100–110. https://doi.org/10.5430/jnep.v6n5p100
nomenological analysis: Theory, method and research. Sage Wertz, F. J., Charmaz, K., McMullen, L. M., Josselson, R., Anderson,
Publications. R., & McSpadden, E. (2011). Five ways of doing qualitative
Švab, I., Klemenc-Ketiš, Z., & Zupanič, S. (2023). New challenges in analysis: Phenomenological psychology, grounded theory,
scientific publications: Referencing, artificial intelligence, and discourse analysis, narrative research, and intuitive inquiry.
ChatGPT. Zdravstveno Varstvo, 62(3), 109–112. https://doi.org/ Guilford Press.
10.2478/sjph-2023-0015 Whetten, D. A. (1989). What constitutes a theoretical contribution?
Sweeney, A., Greenwood, K. E., Williams, S., Wykes, T., & Rose, The Academy of Management Review, 14(4), 490–495. https://
D. S. (2013). Hearing the voices of service user researchers in doi.org/10.2307/258554
collaborative qualitative data analysis: The case for multiple Zhang, H., Wu, C., Xie, J., Lyu, Y., Cai, J., & Carroll, J. M. (2023)
coding. Health Expectations: An International Journal of Public Redefining qualitative analysis in the AI era: Utilizing ChatGPT
Participation in Health Care and Health Policy, 16(4), e89–e99. for efficient thematic analysis. arXiv preprint arXiv:
https://doi.org/10.1111/j.1369-7625.2012.00810.x 2309.10771.