Why a separate discussion of method and ethics?
Large language models are increasingly used to read interview transcripts, suggest codes, summarize themes and draft reports. These tools can save time, but interpretation, sensitivity to context and researcher accountability cannot be delegated to a tool. Two questions therefore need to be kept apart: the methodological question (does AI fit the logic of this analysis?) and the ethical and legal question (what conditions must be met before participant data is sent to a third-party service?).
This guide handles both questions together but under separate headings: where AI can help, which risks stand out, what publication ethics bodies and Turkish institutions expect, and which data protection points you should check with your institution. Legal topics are presented here not as definitive conclusions but as questions to raise with your institution's data protection officer or legal counsel.
Where AI can help and where it cannot
Morgan (2023) tested ChatGPT on two datasets that researchers had already analyzed and reported that the tool did relatively well at identifying concrete, descriptive themes but fell short on subtler themes that require interpretation. This suggests a practical division of labor: AI can assist with mechanical and descriptive work, while making meaning and interpreting remain the researcher's job.
| Task | Reasonable assistance | Must stay with the researcher |
|---|---|---|
| First reading and preparation | Suggesting consistent segment and speaker markup in long transcripts | Deciding what counts as data and how context is preserved |
| Coding | Suggesting candidate codes and excerpts based on the existing codebook | Verifying that a suggestion truly fits the code's definition and rules |
| Theme development | Suggesting draft groupings of code clusters | Deciding what a theme means and how it relates to the research question |
| Reporting | Outline drafts and language editing suggestions | Responsibility for interpretation, inferences, limitations and quote selection |
| Quality checks | Drawing attention to segments that may have been missed | Deliberately searching for negative cases and contradictions |
The “reasonable assistance” column is not a permission list. Which tasks can be left to a tool depends on your method, the scope of your ethics approval and the sensitivity of your data. You can find the basic steps of coding in our interview coding guide.
Epistemological fit: what do reflexive approaches say?
Qualitative research is not a single tradition. In approaches that work with a predefined codebook and value coding consistency, having AI generate candidate suggestions against the codebook is relatively compatible with the logic of the method. Reflexive approaches present a different picture.
Braun and Clarke (2019) emphasize that in reflexive thematic analysis themes do not simply “emerge” from the data but are generated by the researcher, and that researcher subjectivity is an analytic resource rather than an error to be removed. In this framework, coding is the product of a long and deep engagement with the data. If ready-made code or theme suggestions lead a researcher to approve them before building that engagement, the method itself is weakened.
Main risks
The risks of AI-assisted analysis are not limited to technical errors; some of them quietly change how researchers work.
- Hallucinated quotes: Language models can produce fluent, plausible sentences that do not exist in the source. Letting an unmatched quote into a report means attributing words to a participant that they never said.
- Decontextualized quotes: Even a verbatim quote can reverse its meaning if it is selected without reading what comes before and after; irony, reported speech and repeating an interviewer's question are typical cases.
- Bias: Models can reflect patterns in their training data, making certain ways of speaking, regional expressions or minority experiences less visible or misclassifying them.
- Over-reliance: Suggestions that look fast and tidy can loosen critical reading. Over time, approving can replace analyzing.
- Loss of reflexivity: When researchers work from suggestions without building their own relationship with the data, they have fewer chances to notice and question their own assumptions.
- Traceability: The same prompt can produce different outputs at different times, which makes a record of how each result was produced essential.
A researcher conducting interviews on remote work sees the following quote suggested for the code “loneliness”: “Working from home made me completely isolated.” Opening the transcript, she finds that the participant actually said: “Everyone says working from home makes people isolated, but that did not happen to me.” The suggested sentence does not appear verbatim in the source, and the context says the opposite. The researcher rejects the suggestion, reconsiders the segment as “experience contradicting expectations” and records the decision in a reflexive memo.
Accountability, transparency and authorship
However much AI is used, the researcher remains responsible for the accuracy of findings, the fidelity of quotes to their source and the interpretations offered. The main publication ethics texts converge on this point:
- COPE (2023): AI tools cannot be listed as authors because they cannot take responsibility for the work, cannot declare conflicts of interest and cannot manage copyright and license agreements. Authors who use such tools for writing, data collection or analysis should disclose which tool was used and how in the methods or a similar section, and they remain responsible for all content, including parts produced with the tool.
- ICMJE (2025): Recommends that journals ask at submission whether AI-assisted technologies were used. Use for writing assistance should be described in the acknowledgments, and use for data collection, analysis or figure generation in the methods. Chatbots should not be listed as authors because they cannot be held responsible for the accuracy, integrity and originality of the work.
- UNESCO (Miao & Holmes, 2023): Advocates a human-centered approach to generative AI in education and research, highlighting the protection of human agency, attention to data privacy, and institutional evaluation of tools linked to clear policies.
A fictional disclosure for a methods section: “Interview transcripts were uploaded to the analysis software after names and identifying details were replaced with participant codes. The codebook was developed by the researchers. In the second coding round, code suggestions offered by the software through a language model provider were used only to identify segments that might have been missed (tool name, version and dates of use are stated here). Each suggestion was compared with the source text and its context and then accepted, modified or rejected. Themes and interpretations are the researchers' own. Participants were informed that transcripts might be processed by a provider operating outside Türkiye, and this use was described in the ethics committee application.” Your actual disclosure should reflect your own process.
Institutional guidance in Türkiye
The Council of Higher Education (Yükseköğretim Kurulu, 2024) has published an ethics guide on the use of generative AI in scientific research and publication activities at higher education institutions. The guide sets out principles for transparent and responsible use of these tools; clear disclosure of use and the researcher's continued responsibility for generated content are central to its approach. If you work at a university, also check whether your institution has issued its own directive based on this guide.
TÜBİTAK (2025) has published a guide on the responsible and trustworthy use of generative AI in its support processes. Its scope is application processes for TÜBİTAK support programs; it should not be read as a general rulebook governing the analysis stage of every study. If your project is funded by TÜBİTAK or you are preparing an application, consult the current text directly to see which roles and stages it covers.
Data protection: information, pseudonyms and transfers
Interview transcripts often contain personal data: names, workplaces, places of residence, accounts of health or beliefs. Sending that text to an AI provider opens the data to a new processing activity and, in most cases, a new recipient. The points below are considerations to check with your institution's data protection officer or legal counsel.
- Information and consent: Does the participant information sheet clearly say that transcripts may be processed by an AI service, who the provider is and whether data will be transferred abroad? Consent under research ethics and the duty to inform under data protection law are different obligations; having one does not automatically satisfy the other.
- Legal basis: Document with your institution which processing condition in Law No. 6698 the processing relies on and, where special categories of personal data (such as health data) are involved, how the additional conditions are met.
- Pseudonymization: Before sending text to a third-party provider, replace names, organization names and easily identifying details with participant codes; keep the key list separately with restricted access. Pseudonymized data may remain personal data as long as re-identification is possible, so pseudonyms do not remove your other obligations.
- Data minimization: Do not send segments the analysis does not need (introductions, off-the-record chat).
- Roles and contracts: Check who acts as data controller and data processor, and which commitments the platform and its sub-processors operate under. This platform's sub-processors are listed on the sub-processors page, and its technical and organizational measures on the security page.
If the provider processes data outside Türkiye, the rules on cross-border transfers may apply. Article 9 of Law No. 6698 was amended in 2024, and a transfer must in any case also rest on one of the processing conditions in the Law. The current structure of the article can frame your conversation with your institution:
| Topic | In brief | Question for your institution |
|---|---|---|
| Adequacy decision | An adequacy decision by the Board concerning the destination country, sector or international organization | Is there an adequacy decision in force for where the provider processes the data? |
| Appropriate safeguards | Without an adequacy decision, safeguards such as standard contracts, provided that data subjects can exercise their rights and have effective remedies; a standard contract is notified to the Authority within five business days of signing | Which safeguard is used? Has a standard contract been signed and notified within the deadline? |
| Incidental transfers | Where neither an adequacy decision nor appropriate safeguards exist, transfers that may be made only in limited cases and on an incidental basis | Is this transfer truly incidental, or is it regular and ongoing processing? |
The Turkish Personal Data Protection Authority (Kişisel Verileri Koruma Kurumu, 2025) has published a question-and-answer guide on generative AI and personal data protection. It is worth reviewing when you prepare your internal policy or participant information sheet.
Ethics committees, a checklist and the audit trail
If your ethics committee application did not anticipate AI use or data transfers to a third-party provider, ask the committee before starting such use whether it requires an amendment or a new approval. The scope of the approval should be consistent with what participants were told and with what you actually do.
- Specify in writing at which stage and for what purpose AI will be used, alongside your research question and approach.
- Confirm that your ethics committee application and participant information sheet cover this use.
- Check the legal basis, any cross-border transfers and the relevant contracts with your data protection officer.
- Read the AI policies of your university, your funder and your target journal.
- Pseudonymize transcripts before sending them and remove segments the analysis does not need.
- Accept, modify or reject each suggestion only after comparing it with the source text and its context.
- Reread AI-written narrative text (theme definitions, summaries, report drafts) and revise it with your own interpretation.
- Record in reflexive memos how suggestions influenced your thinking.
- In your methods section, disclose the tool's name, version, dates of use, purpose and how human oversight was carried out.
The audit trail is the evidence behind this checklist. Keeping a record of which tool ran on which data and when, which suggestions were accepted, modified or rejected, and the rationale for merging or splitting codes lets you answer questions from reviewers and ethics committees and revisit your own analytic decisions later. If you code as a team, these records also help in discussions of intercoder agreement.
How YouReply Qualitative supports this process
In YouReply Qualitative, AI is optional and controlled per study with a switch. When a new study is created, the switch currently starts on; you can turn it off at creation or later in study settings. While it is off, no new AI runs are started. The provider is OpenAI (API).
- Data sent: When AI is on, transcript segment text (including speaker labels), codebook names, definitions and rules, and the research question are sent. Theme, pattern and report suggestions send code names, short excerpts and participant codes.
- No automatic pseudonymization: Speaker labels are sent as written. We therefore recommend using participant codes rather than real names as speaker labels and pseudonymizing transcripts before upload.
- Suggestions stay pending: Each suggestion waits until you accept it, edit it by changing its code or reject it; unaccepted suggestions never enter counts, matrices or reports.
- Quotes must match the source: A suggested quote can be accepted only if it matches the source text (differences in case, whitespace and quotation marks are tolerated), and the stored excerpt is taken from the source. AI-written theme descriptions, pattern summaries and report drafts require your review.
- Records: Each coding records its origin (manual, AI suggestion accepted, or accepted with a different code); codebook operations are logged, and merges and splits can carry a rationale note. You can keep analytic, methodological, reflexive and general memos.
According to the provider's published data controls documentation, API data is not used to train models unless the organization opts in, abuse-monitoring logs are retained for up to 30 days by default, and zero data retention requires OpenAI's approval. These are the provider's published terms and should not be read as guarantees about our configuration.
Summary
- AI can help with descriptive and mechanical work; responsibility for interpretation, context and findings stays with the researcher.
- In reflexive approaches, explain AI use in the methods section and justify how subjectivity and interpretive responsibility were protected.
- According to COPE and ICMJE, AI cannot be an author; disclose which tool was used and for what purpose.
- Pseudonymize transcripts before sending them, and check information duties, legal basis and cross-border transfers with your data protection officer.
- Turning AI off does not by itself satisfy ethics committee or data protection obligations.
- Keep an audit trail of accepted, modified and rejected suggestions together with your reflexive memos.
References
- Braun, V., & Clarke, V. (2019). Reflecting on reflexive thematic analysis. Qualitative Research in Sport, Exercise and Health, 11(4), 589–597. https://doi.org/10.1080/2159676X.2019.1628806
- Committee on Publication Ethics. (2023). Authorship and AI tools [COPE position statement]. https://publicationethics.org/guidance/cope-position/authorship-and-ai-tools
- International Committee of Medical Journal Editors. (2025). Recommendations for the conduct, reporting, editing, and publication of scholarly work in medical journals: Artificial intelligence (AI) use by authors. https://www.icmje.org/recommendations/browse/artificial-intelligence/ai-use-by-authors.html
- Kişisel Verileri Koruma Kurumu. (2025). Üretken yapay zekâ ve kişisel verilerin korunması rehberi (15 soruda) [Guide on generative AI and the protection of personal data (in 15 questions)]. https://www.kvkk.gov.tr/Icerik/8547/uretken-yapay-zeka-ve-kisisel-verilerin-korunmasi-rehberi-15-soruda
- Miao, F., & Holmes, W. (2023). Guidance for generative AI in education and research. UNESCO. https://doi.org/10.54675/EWZM9535
- Morgan, D. L. (2023). Exploring the use of artificial intelligence for qualitative data analysis: The case of ChatGPT. International Journal of Qualitative Methods, 22. https://doi.org/10.1177/16094069231211248
- TÜBİTAK. (2025). Destek süreçlerinde üretken yapay zekânın (ÜYZ) sorumlu ve güvenilir kullanımı rehberi [Guide to the responsible and trustworthy use of generative AI in support processes]. https://tubitak.gov.tr/sites/default/files/2025-10/UYZ_Rehberi_v03_TR.pdf
- Yükseköğretim Kurulu. (2024). Yükseköğretim kurumları bilimsel araştırma ve yayın faaliyetlerinde üretken yapay zekâ kullanımına dair etik rehber [Ethics guide on the use of generative AI in scientific research and publication activities of higher education institutions]. https://www.yok.edu.tr/tr/news/yuksekogretim-kurulu-yuksekogretim-kurumlari-bilimsel-arastirma-ve-yayin-faaliyetlerinde-uretken-yapay-zeka-kullanimina-dair-etik-rehber-hazirladi-zisSm
- 6698 sayılı Kişisel Verilerin Korunması Kanunu. (2016, 2024 değişiklikleriyle). Resmî Gazete. [Law No. 6698 on the Protection of Personal Data, as amended in 2024] https://www.mevzuat.gov.tr/mevzuat?MevzuatNo=6698&MevzuatTur=1&MevzuatTertip=5