03 · What You Need to Know
How AI Assistance Affects Qualitative Analysis and Its Reporting
What Counts as AI-Assisted Qualitative Analysis?
AI-assisted qualitative analysis involves using artificial intelligence to support the organization, examination, coding, comparison, synthesis, or interpretation of nonnumerical research materials.
These materials may include interview transcripts, focus group discussions, open-ended survey responses, field notes, institutional documents, or online discussions.
AI systems may suggest codes, identify patterns, summarize responses, compare participant accounts, or generate preliminary interpretations.
However, these activities are not interchangeable. Organizing excerpts according to predefined categories differs from developing an interpretive account of participants' experiences.
The distinction becomes important when researchers describe how their findings were produced.
Why Is AI Assistance Particularly Consequential in Qualitative Research?
Qualitative analysis frequently requires researchers to consider context, ambiguity, relationships, and the meanings participants attach to their experiences.
A generative AI system may produce a coherent thematic summary without adequately representing contradictions or the circumstances in which statements were made.
For example, students who describe using AI because of demanding academic workloads may be interpreted as expressing technological enthusiasm, even when their accounts primarily concern institutional pressure.
Such interpretations may sound plausible while overlooking important distinctions.
AI assistance is not necessarily inappropriate, but its role should be evaluated in relation to the epistemological assumptions and analytical procedures of the study.
Researchers must determine whether the generated output supports, distorts, or prematurely simplifies their engagement with the evidence.
What Do Major Publishing Authorities Say About AI-Assisted Analysis?
The International Committee of Medical Journal Editors (ICMJE) recommends disclosure of AI-assisted technologies used in submitted research and identifies AI involvement in data analysis as appropriate for methods reporting.
The Committee on Publication Ethics (COPE) similarly emphasizes transparency concerning AI tools used in research activities and maintains that human authors remain responsible for submitted work.
These positions support disclosure of consequential AI-assisted qualitative analysis, although they do not establish one universal procedure for conducting or validating qualitative research.
Researchers should therefore follow both the journal's AI-use policy and the methodological standards associated with their analytical approach.
The broader requirements concerning which AI activities require disclosure should be interpreted in relation to what the system actually contributed.
Is AI-Assisted Coding the Same as AI-Assisted Qualitative Analysis?
Not necessarily. Coding may form part of qualitative analysis, but qualitative inquiry often extends beyond assigning labels to excerpts.
For example, a researcher may use AI to classify responses into predefined categories such as accessibility, instructor support, or technical difficulties.
Another researcher may ask AI to identify underlying meanings, construct themes, explain contradictions, and develop an interpretive narrative.
The second activity gives AI a broader analytical role.
AI-assisted coding
Using AI to suggest or assign codes to research materials according to an established or developing analytical framework.
AI-assisted interpretation
Using AI to propose meanings, relationships, themes, or explanations that contribute to the research findings.
Both activities may require disclosure, but the description should distinguish classification from interpretation.
The specific question of disclosing AI-assisted research data coding concerns one component of a potentially broader qualitative workflow.
Does Using ChatGPT to Generate Themes Require Disclosure?
When ChatGPT materially contributes to theme development, methodological disclosure is generally appropriate and may be explicitly required by the journal.
For example, a researcher may ask the system to examine interview transcripts and suggest themes concerning students' perceptions of AI-assisted learning.
If the generated themes influence the final analysis, the researcher should explain how the suggestions were produced, reviewed, modified, or rejected.
However, the researcher should not assume that an AI-generated list of topics constitutes a completed thematic analysis.
The distinction between recurring topics and analytically developed themes depends on the methodological approach.
A theme may need to express a meaningful pattern relevant to the research question rather than merely summarize a frequently mentioned subject.
Can AI Perform Reflexive Thematic Analysis?
AI may assist with activities within a reflexive thematic analysis workflow, but claiming that an AI system independently performed reflexive thematic analysis raises methodological concerns.
Braun and Clarke's approach emphasizes the researcher's active role in generating codes, developing themes, and interpreting meaning.
Within this framework, themes are not simply discovered as objective entities waiting inside the dataset. They are developed through situated analytical engagement.
An AI-generated thematic structure therefore cannot automatically substitute for the interpretive work required by the approach.
Researchers may examine AI suggestions critically, but they should explain how those suggestions were incorporated without misrepresenting the methodology.
Calling an automated summary "reflexive thematic analysis" merely because it contains theme labels would be difficult to defend.
Does AI-Human Coding Agreement Establish Qualitative Credibility?
Not automatically. Agreement statistics may be useful in some coding designs, particularly those involving predefined categories and reliability-oriented procedures.
However, their appropriateness depends on the analytical framework.
In reflexive thematic analysis, coding differences may reflect interpretive perspectives rather than errors requiring numerical reconciliation.
A high agreement score between AI and a human coder does not necessarily establish that the categories are meaningful, theoretically appropriate, or sensitive to participants' experiences.
Researchers should avoid treating inter-rater agreement as a universal requirement for qualitative rigor.
Verification must be consistent with the epistemological and methodological assumptions of the study.
What Is the Difference Between AI-Assisted Thematic Analysis and Automated Topic Modeling?
Automated topic modeling and qualitative thematic analysis may both identify patterns in textual data, but they are not methodologically equivalent.
Topic modeling typically uses computational procedures to identify statistical patterns of word co-occurrence or related textual representations.
Qualitative thematic analysis involves interpreting patterns of meaning in relation to the research question and analytical approach.
AI-generated topics may provide useful exploratory material, but their existence does not automatically establish interpretive themes.
Researchers should describe the actual procedure performed rather than presenting computationally generated topics as though they emerged through a different qualitative methodology.
Can AI-Assisted Analysis Introduce Interpretive Bias?
Potentially. AI systems may favor familiar explanations, standardize unusual expressions, or overlook accounts that contradict dominant patterns.
For example, a system analyzing student interviews may repeatedly categorize dissatisfaction with online learning as poor motivation while overlooking structural barriers such as internet connectivity or inaccessible course materials.
These errors could alter how participants' experiences are represented.
Researchers should examine negative cases, contextual differences, and interpretations that do not fit the generated categories.
However, bias should be evaluated empirically and contextually rather than presumed to be identical across all AI systems or datasets.
Human researchers also bring interpretive assumptions to qualitative analysis, making reflexivity relevant whether or not AI is used.
Does AI Assistance Remove the Need for Researcher Reflexivity?
No. Reflexivity concerns how researchers' positions, assumptions, experiences, and decisions influence the production of knowledge.
AI assistance introduces additional analytical influences that may themselves require critical examination.
For example, researchers may consider why they accepted one generated interpretation while rejecting another, or how their prompts directed the system toward particular categories.
The apparent neutrality of an automated response should not be mistaken for the absence of assumptions.
Prompts, model behavior, input selection, and human interpretation all influence the analytical process.
Reflexive documentation may therefore help explain how AI-assisted suggestions interacted with the researcher's own judgments.
Should AI Be Used to Analyze Entire Interview Transcripts?
Whether this is appropriate depends on the research design, data protections, analytical approach, and tool capabilities.
Analyzing complete transcripts may preserve more context than processing isolated excerpts, but it can also introduce confidentiality concerns and practical limitations.
AI systems may omit portions of lengthy materials, misunderstand speaker relationships, or generate summaries that overemphasize certain passages.
Researchers should establish that the tool can process the relevant materials appropriately and that its use is authorized.
They should also verify consequential interpretations against the original transcripts.
Uploading an entire dataset to an AI service is not automatically methodologically preferable to a more controlled analytical workflow.
Does AI-Assisted Translation Affect Qualitative Analysis Disclosure?
Yes, when translation forms part of the analytical process.
For example, a researcher may translate Filipino interview transcripts into English before asking an AI system to identify codes or themes.
The resulting analysis may be influenced by both the translation and the subsequent AI-assisted interpretation.
Culturally specific expressions may lose important nuances during translation, affecting the categories assigned to participants' accounts.
Researchers should therefore document consequential translation and analytical procedures separately where appropriate.
The requirements concerning AI-assisted translation in research become particularly relevant when translated materials are used as analytical evidence.
What Privacy and Ethics Issues Arise?
Qualitative datasets may contain identifiable information, sensitive experiences, institutional details, or narratives that permit indirect identification.
Uploading transcripts to an external AI service may create risks involving data retention, access, confidentiality, and compliance with participant consent.
Researchers should verify that the intended processing is authorized under the ethics approval, institutional policies, applicable law, and service arrangements.
Removing names alone may not adequately de-identify detailed qualitative accounts.
Where necessary, institutionally approved tools or more restricted processing arrangements may be appropriate.
Disclosure in a published manuscript does not retrospectively authorize an impermissible transfer of participant information.
What Should an AI-Assisted Qualitative Analysis Disclosure Include?
The statement should accurately explain the AI system's analytical role and the procedures through which researchers evaluated its outputs.
| Reporting Element |
Purpose |
| AI tool or model |
Identifies the technology used for analytical assistance. |
| Materials processed |
Explains whether AI examined transcripts, documents, excerpts, or coded data. |
| Analytical approach |
Establishes the methodological framework governing interpretation. |
| AI function |
Distinguishes summarization, coding, theme suggestions, and interpretation. |
| Human analytical involvement |
Explains how researchers developed, revised, or rejected AI outputs. |
| Verification and reflexivity |
Describes relevant checks and the consideration of interpretive influences. |
| Limitations |
Identifies uncertainties arising from the AI-assisted procedure. |
Additional details concerning prompts, model versions, or settings may be relevant when required by the journal or necessary to understand the procedure.
The reporting should be proportionate to the role of AI rather than a generic statement that human oversight occurred.
Where Should AI-Assisted Qualitative Analysis Be Disclosed?
Consequential AI-assisted qualitative analysis generally belongs in the methods section because it concerns how research findings were developed.
A journal may also require a designated AI-use declaration.
For example, a study may explain AI-assisted preliminary coding under data analysis while providing a concise disclosure identifying the tool elsewhere in the manuscript.
The statements should be consistent and should not imply that the entire analysis was conducted manually if AI materially contributed to its development.
The distinction between methods reporting and acknowledgment of AI assistance is especially important when the tool influenced interpretation rather than language presentation.
What Should an AI-Assisted Qualitative Analysis Statement Say?
Illustrative Methods Disclosure
AI Assistance During Preliminary Qualitative Coding
"ChatGPT (OpenAI) was used to suggest preliminary codes for selected de-identified interview excerpts. The research team examined these suggestions against the original transcripts, revised or rejected codes where necessary, and developed the final themes through researcher-led interpretation. AI-generated suggestions were not treated as independently validated findings."
This statement is appropriate only if the described procedure actually occurred and the processing was authorized.
Researchers should not claim researcher-led theme development or comprehensive transcript review unless those activities were performed.
The journal may require additional information concerning the model, prompts, and analytical procedures.
Does Disclosure Establish the Trustworthiness of AI-Assisted Findings?
No. Disclosure supports transparency but does not independently establish qualitative credibility.
A researcher may accurately describe AI-generated themes while failing to demonstrate that those themes are supported by the data.
Conversely, a carefully justified analysis may still require disclosure under the journal's policy.
Trustworthiness must be evaluated according to the study's analytical framework, evidence, reflexive practices, and methodological coherence.
Researchers should avoid treating AI-generated agreement, fluent explanations, or repeated themes as sufficient evidence of interpretive validity.
Should Researchers Preserve AI-Assisted Analytical Records?
Relevant records may help researchers reconstruct how AI suggestions influenced coding and interpretation.
These can include prompt templates, generated outputs, analytical memos, revised coding frameworks, theme-development notes, and documented decisions to reject suggestions.
However, the appropriate records depend on the methodology and applicable retention requirements.
Qualitative audit trails should not be reduced to a collection of AI transcripts without explanation of the researcher's decisions.
Researchers must also protect participant confidentiality when preserving prompts or generated responses containing research data.
The broader issue of maintaining records of AI use should be considered in relation to the analytical significance of the assistance.
What If the Journal Does Not Specifically Address AI Qualitative Analysis?
Researchers should consult the journal's broader AI-use policy and the reporting standards relevant to their methodology.
Even where no dedicated AI rule exists, the methods section should accurately explain consequential analytical procedures.
If uncertainty remains, the researcher may seek clarification from the editorial office.
The inquiry should describe whether AI performed preliminary coding, theme generation, summarization, or substantive interpretation.
Policy silence does not eliminate the need to report methods accurately or justify the credibility of the analysis.
