Generative AI in Qualitative Data Analysis: Introducing the Guided AI Thematic Analysis (GAITA) framework

[email protected] Nguyen-Trung [email protected] Kien Nguyen-Trung, PhD GEDSI and Research Lead, Water Sensitive Cities Australia, Monash University The IASSIST Professional Development Committee and the Qualitative Social Science & Humanities Data Interest Group (QSSHDIG) (21 November, 2025) Generative AI in Qualitative Data Analysis: Introducing the Guided AI Thematic Analysis (GAITA) framework Give us feedback via https://forms.gle/1tvHSJh1inviLKAt7 [email protected] Nguyen-Trung [email protected] Structure ■The rise of generative AI in qualitative data analysis ■A Guided-AI Thematic Analysis (GAITA) in the thematic analysis family ■GAITAI: a step-by-step guide with a worked example [email protected] Nguyen-Trung [email protected] The rise of generative AI in qualitative data analysis: current trends Silver, C. (2023) What’s a foot in the Qualitative AI space? (n.d.). Methodspace. Retrieved August 25, 2023, from https://www.methodspace.com/blog/whats-a-foot-in-the-qualitative-ai-space Friese, S. (2025). Conversational Analysis with AI -CA to the Power of AI: Rethinking Coding in Qualitative Analysis. SSRN. https://ssrn.com/abstract=5232579 or http://dx.doi.org/10.2139/ssrn.5232579 For the comparison between three scenarios, see: Nguyen-Trung, K., & Nguyen, N. L. (2025, under review). Narrative-Integrated Thematic Analysis (NITA): AI-Supported Theme Generation Without Coding. https://doi.org/10.31219/osf.io/7zs9c_v1 What are their potentials? Generative AI in qualitative data analysis: Three scenarios Atlas.ti: Conversational AI MaxQDA: AI Assist NVivo: AI Assistant See Friese’s (2025) Conversational Analysis with AI Scenario 1: Integrating AI into existing CAQDAS-packages Scenario 2: Developing New AI Qualitative Research apps Scenario 3: Using AI chatbots (generalpurpose AI) [email protected] Nguyen-Trung [email protected] Generative AI in qualitative data analysis: Current applications Applications Innovating coding approaches Shifting to non-coding approaches -a new paradigm? ‘Query-based approach’ (QBA) (Morgan, 2025) Conversational Analysis with AI (CAAI) (Friese, 2025) ‘Narrative-Intergrated Thematic Analysis’ (NITA) (Nguyen-Trung & Nguyen, 2025) •Reflexive thematic analysis (Hamilton et al., 2023, Drápal et al., 2023; De Paoli, 2024; Prescott et al., 2024) •Content analysis (Chew et al., 2023), •Grounded theory (Sinha et al., 2024; Wachinger et al., 2024) •GAITA, adapted version of Template Analysis (Nguyen-Trung, 2025) “if manual coding of data segments has been the dominant paradigm for qualitative analysis, then yes, I believe that querying with ChatGPT does threaten that dominance.” (Morgan, 2023, p. 9) Friese, S. (2025). Generative AI: A Catalyst for Paradigmatic Change in Qualitative Data Analysis. In Artificial Intelligence (AI) in Social Research. CAB International. Morgan, D. L. (2023). Exploring the Use of Artificial Intelligence for Qualitative Data Analysis: The Case of ChatGPT. International Journal of Qualitative Methods, 22, 16094069231211248. https://doi.org/10.1177/16094069231211248 “This represents a significant shift in beliefs about how researchers can effectively comprehend and interpret reality.” (Friese, 2025, p. 12, italics are ours) [email protected] Nguyen-Trung [email protected] Introduction to Guided AI Thematic Analysis (GAITA) [email protected] Nguyen-Trung [email protected] Nguyen-Trung, K. (2025) ChatGPT in thematic analysis: Can AI become a research assistant in qualitative research?. Qual Quant. https://doi.org/10.1007/s11135-025-02165-z [email protected] Nguyen-Trung [email protected] 2. Preparation 1. Planning PERFECT Reflexivity Continuously PERFECTING •Get familiar with your data •Generate different forms of summaries •Visualise common topics •Code the first transcripts (subset) to form a list of code •Apply the list of codes to the rest of the transcripts •Form a revised list of codes •Groups codes into meaningful clusters to form an initial template •Modify the templates by generating subcodes (hierarchical organization) •Finalize the template by recoding & recategorizing Purposefully plan initial PERFECT procedure to guide GenAI analysis -Prepare the dataset -Select suitable AI tools -Prepare the system with ROLE & GUIDES -Develop Custom GPT / project A Guided AI Thematic Analysis (GAITA): Adapted Adapted from Nguyen-Trung, K., & Nguyen, N. L. (2025, under review). Narrative-Integrated Thematic Analysis (NITA): AI-Supported Theme Generation Without Coding. https://doi.org/10.31219/osf.io/7zs9c_v1 •Generate themes based on the final template •Edit and refine themes based on your interpretations •Generate an overarching theme •Write-up GAITA: A step-by-step guide with a worked example [email protected] Nguyen-Trung [email protected] Stage 2: Preparing data and GenAI system •Preparing the dataset •Anonymize or pseudonymize •Prepare 11 transcripts in pdf format with no headings (to avoid domain-led theme development). [email protected] Nguyen-Trung [email protected] Stage 2: Preparing data and GenAI system (2) •Preparing the GenAI system •turning off data for training •setting up in ChatGPT system involving the use of ROLE and GUIDES: The GUIDES framework ensures ChatGPT’s behaviors adhere to research ethics and quality through Guidelines (G), Understanding (U), Integrity (I), Document Referencing (D), Expression (E), and Suggestions (S). The ROLE framework covers: Role (R), Objectives (O), Lexicon (L), and Ethics (E), focusing on introducing our research roles, goals, jargons, and ethical standards. Nguyen-Trung, K., & Nguyen, N. L. (2025, under review). Narrative-Integrated Thematic Analysis (NITA): AI-Supported Theme Generation Without Coding. https://doi.org/10.31219/osf.io/7zs9c_v1 [email protected] Nguyen-Trung [email protected] Stage 2: Preparing data and GenAI system (3): Custom GPT •Preparing the GenAI system (cont.) •creating a customized GPT ‘Guided AI Thematic Analysis); •starting training the GPT by uploading 11 transcripts or the single file containing them; •Where you instruct GAITA (this can include actor – assigning a role) •Where you upload transcripts Nguyen-Trung, K., & Nguyen, N. L. (2025, under review). Narrative-Integrated Thematic Analysis (NITA): AI-Supported Theme Generation Without Coding. https://doi.org/10.31219/osf.io/7zs9c_v1 [email protected] Nguyen-Trung [email protected] Stage 2: Preparing data and GenAI system (4): Projects •Preparing a GenAI system by using Projects feature: •creating a Project named “GAITA approach”; •starting training the GPT by uploading the dataset ‘Down East’ and the original papers and applying GUIDES •Where you instruct GAITA (including actor) •Where you upload transcripts Nguyen-Trung, K., & Nguyen, N. L. (2025, under review). Narrative-Integrated Thematic Analysis (NITA): AI-Supported Theme Generation Without Coding. https://doi.org/10.31219/osf.io/7zs9c_v1 [email protected] Nguyen-Trung [email protected] Stage 2: Preparing data and GenAI system (4): Projects •In Claude •creating a Project named “GAITA approach”; •starting training the GPT by uploading the dataset ‘Down East’ and the original papers and applying GUIDES •Where you instruct GAITA (including actor) •Where you upload transcripts [email protected] Nguyen-Trung [email protected] Step 3: Data familiarization (1) •Purpose: •To become familiar with the data in order to prepare for subsequent in-depth analyses •To verify your understanding through goal-oriented conversations •Expected outcome: •You understand the data and are prepared for deeper analysis [email protected] Nguyen-Trung [email protected] Step 3: Data familiarization (2) •AI’s tasks: •Create different types of summaries •Open prompt: Prompt 3.1: “Can you please provide a summary of this interview [Interview number 1 with Barbara]?” •Closed prompt: Prompt 3.2: “Highlight the key ideas of the interview in three key bullet points.” •Create a whole dataset summary •Prompt 3.3: “Can you please provide me with a 400word summary of key ideas from all the interviews in the uploaded file? All interviews are numbered, and each starts with a short introduction of the interviewees.” [email protected] Nguyen-Trung [email protected] Step 3: Data familiarization (3) [email protected] Nguyen-Trung [email protected] Step 4: Preliminary coding •Purpose: •To assign labels/codes to the relevant data •Expected outcome: •An initial list of codes •A revised list of codes after coding all transcripts •A list of codes that captures nuances related to the questions [email protected] Nguyen-Trung [email protected] Step 4: Preliminary coding (2) •AI’s tasks: •Coding the first transcript(s) •Prompt 4.1: •You’re a qualitative research assistant [actor]. •You will help me identify the relevant codes [task] from the following transcripts in response to ‘How do these people connect to Down East?’. [context: research question] •Codes are labels that assign summative, salient, essence-capturing meaning to a portion of data. [context: knowledge]. •#The final outputs are a table like this: Column 1: Code; Column 2: Description of code meaning; Column 3: Quotation representing the code from transcript. [output]. •##Here is the transcript [insert the transcript] [reference]. •Check & reflect, write memo [email protected] Nguyen-Trung [email protected] Step 5: Template formation & refinement (2) •AI’s taks: •Prompt 5.1: Form the initial template •“From the following refined list of codes, please group codes into clusters (i.e., more abstract codes) in response to ‘How do these people connect to Down East?’. •# For each cluster, please retain specific codes and quotations [maintain evidence]; •if two or more transcripts share a cluster, group them [avoid AI’s simplification of data]. •##The final output will be a table: Column 1: Cluster; Column 2: Codes; 3. Description of Cluster Meaning; Column 4: Quotation from Transcripts. Please make sure you quote the names of the interviewees for each quote. •###Here is the list of code [paste] [email protected] Nguyen-Trung [email protected] Step 5: Template formation & refinement (3) •AI’s tasks: •Prompt 5.2: Refine the template by developing subcodes •“Now I need you to further develop the coding table above by generating sub-codes for each code based on the quotations I give you [quotes from previous rounds]. •# Task: •(1) place each quotation under a suitable sub-code; •(2) create new sub-codes as needed. •Final table: cluster, code, sub-codes, sub-code description, quotations. •### Here is the first set of quotations under ‘family history’ [paste quotes] [email protected] Nguyen-Trung [email protected] Initial template with 5 clusters Working with each cluster Refine clustering by adding or deleting codes, re-clustering [email protected] Nguyen-Trung [email protected] Final template [email protected] Nguyen-Trung [email protected] Step 6: Theme development •Purpose: •Develop themes to interpret the data and answer the research question •Expected outcomes: •Initial list of themes based on the final template •Revise, name, and further develop meaningful themes through interpretive analysis •Develop overarching themes and build a narrative structure [email protected] Nguyen-Trung [email protected] Step 6: Theme development: Task (1) •AI’s tasks: •Prompt 6.1: •“From the above table of clusters, please generate themes across the clusters and codes in response to ‘How do these people connect to Down East?’. •# Theme = recurrent, distinctive features of participants’ accounts linked to context. •##The final output will be a table: Column 1: Theme; Column 2: Clusters and Codes used for the theme; 3. Description of Theme Meaning; Column 4: Quotation from Transcripts. •Please make sure you quote the names of the interviewees for each quote. [email protected] Nguyen-Trung [email protected] Final themes [email protected] Nguyen-Trung [email protected] Nguyen-Trung, Kien, Documenting debates on GenAI in (reflexive) qualitative research (November 15, 2025). Available at SSRN: https://ssrn.com/abstract=5750283 or h ttp://dx.doi.org/10.2139/ssrn.5750283 To use or not to use? [email protected] Nguyen-Trung [email protected] GAITA’s position in thematic analysis family (1) •Is GAITA reflexive? Lupton, D. (2025, June). This researcher posed the question – ‘Is ChatGPT useful and reliable for qualitative data thematic analysis?’. The answer, unsurprisingly, is [Status update]. LinkedIn. Retrieved June 22, 2025, from https://www.linkedin.com/posts/deborah-lupton7ab43b270_chatgpt-in-thematic-analysis-can-aibecome-activity-7338328585570537472-jG3j Themes as shared patterns of meanings Themes as topic summaries [email protected] Nguyen-Trung [email protected] Theme development: GAITA’s position (2) •NITA/GAITA is an in-between approach Lupton, D. (2025, June). Is GAITA/NITA reflexive TA? NITA/GAITA inherits from reflexive TA (Braune & Clarke) but is based on pragmatism