Short answer: An ethics application for AI-moderated interviews must explain four things a standard interview protocol never had to: what the AI does in the conversation, which providers process participants' words and where, how distress and disclosure are handled without a human in the room, and how AI outputs are checked before they become findings. Answer each one in its own section and attach evidence for it.
Last updated: November 10, 2026
Writing an ethics application for AI-moderated interviews is now a routine task for PhD students and research labs, and it is still the step where many studies stall. Ethics committees and IRBs know how to review a human interviewer with a topic guide. When an AI asks the follow-up questions, transcribes the audio and drafts the first codes, they ask new questions, and a vague answer usually comes back as a request for revisions.
This guide gives you a section-by-section template, a map of the questions reviewers ask, and boilerplate wording you can adapt. It is written for researchers in Europe and elsewhere who submit to a university research ethics committee (REC) or an institutional review board (IRB). It does not replace your institution's own forms or your data protection officer's advice.
Key takeaways
- Committees increasingly treat an AI tool that talks to participants as part of the research procedure, so describe it the way you would describe a human interviewer: role, script, limits.
- The most common gap is the data-flow description. Name each processor, the data it receives, where it is processed and how long it is kept.
- A DPIA is not automatic for every interview study, but sensitive topics at scale can trigger one under GDPR Article 35. Decide early and record the reasoning.
- Plan for distress without a human in the room: stop rules, a visible exit, and a named person who follows up.
- Say how a human checks AI transcription, follow-up questions and coding before anything is reported.
Why AI-moderated interviews change the ethics application
An AI moderator changes three things a committee reviews: who interacts with the participant, who processes the data, and who produces the first interpretation. Each change needs its own answer in the application.
University guidance published in 2025 and 2026 points the same way. The University of Tennessee states that an AI tool used to interact with participants or obtain data from them is considered engaged in human subjects research. The University of Michigan's human research protection program asks applicants to explain whether data will be shared with external or third-party AI tools, what will be shared and under what conditions. The University of Auckland's doctoral guidelines, effective 1 September 2026, require intended generative AI use in research with human participants to be detailed in the ethics application.
In Europe, data protection sits alongside ethics review. On 15 April 2026 the European Data Protection Board adopted draft Guidelines 1/2026 on processing personal data for scientific research, put out for public consultation until 25 June 2026. Among other points, the draft addresses how responsibilities are split between controllers, joint controllers and processors, which is exactly the question an AI interview platform raises. As of November 10, 2026 we could not confirm a final version, so cite the draft as a draft.
What reviewers ask, and where to answer it
Most revision requests for AI interview studies come from a short list of questions. The map below is a Qualitati-owned framework built from the institutional guidance cited above. Use it as a pre-submission check: every row should point to a section of your application and a piece of evidence.
| Reviewer question | Where to answer it | Evidence to attach |
|---|---|---|
| Will participants know they are talking to an AI? | Procedure; information sheet | Consent text with the AI disclosure; screenshot of the interview start screen |
| What exactly does the AI do and not do? | Procedure | Interview guide, AI instructions or settings, list of topics it may not raise |
| Who processes the audio and transcripts, and where? | Data management | Data-flow table; vendor privacy policy and data processing agreement |
| Is the data used to train AI models? | Data management | Vendor terms on training and retention, quoted with the date you checked them |
| What happens if a participant becomes distressed? | Risk and mitigation | Stop rules, support resources, named follow-up contact |
| How do you handle disclosures of harm or illegal activity? | Risk and mitigation | Disclosure protocol and the limits of confidentiality in the consent text |
| Can participants withdraw, and what is deleted? | Consent; data management | Withdrawal procedure and the deletion steps on the platform |
| How are AI transcripts and codes checked? | Analysis | Verification plan: sample checks, second coder, audit trail |
| Is a DPIA required? | Data protection | DPIA, or a short screening note explaining why one is not needed |
How to write an ethics application for AI-moderated interviews: section by section
The template below follows the order of a typical REC or IRB form. Rename the headings to match your institution's form, and keep the content.
1. Study summary and the role of AI
State the research question, then one plain sentence on the AI's role. Example wording: "Interviews will be conducted by an AI interviewer that asks the questions in the approved interview guide and generates short follow-up questions based on participants' answers. It does not give advice, does not raise topics outside the guide, and cannot access any information about participants beyond what they type or say in the interview."
2. Procedure
Describe the participant's path step by step: invitation, information sheet, consent, interview (text or voice), debrief. Attach the interview guide and say how follow-up questions are constrained. If you piloted the guide, say so and summarize what changed; our guide on pilot testing an AI interview guide covers what to check.
3. Recruitment and consent
The information sheet should say, in plain language, that an AI conducts the interview, that responses are recorded and transcribed, which organizations process the data, and how to withdraw. Some institutions now ask for explicit consent to AI transcription: the University of Roehampton's September 2025 policy is one example. For the disclosure wording itself, see what informed consent means in AI-moderated research and our note on EU AI Act Article 50.
4. Data management and processors
This is the section committees question most. Write a data-flow table with one row per processor: the interview platform, its AI model providers, the transcription service and your own storage. For each, give the data received, the processing location, the retention period and the legal safeguard for any transfer outside the EEA. Check whether the vendor publishes its sub-processors and whether AI providers retain or train on inputs, and record the date you checked.
Then state your own retention and deletion plan: when raw audio is deleted, when transcripts are pseudonymized, and what is archived. If you plan to share data later, our qualitative data sharing and GDPR guide covers archiving.
5. Data protection impact assessment
GDPR Article 35 requires a DPIA when processing is likely to result in a high risk, and lists large-scale processing of special category data as one example. An interview study on health, sexuality, religion or political views can reach that threshold, especially across many participants. The EDPB adopted a DPIA template in 2026; it is not mandatory, and your institution may have its own. Ask your data protection officer early. If you conclude no DPIA is needed, write down why.
6. Risks and mitigation
List the risks specific to AI moderation and the safeguard for each:
- Distress with no human present: a visible way to pause or end the interview at any time, support resources shown at the end, and a named researcher who reviews flagged sessions within an agreed time.
- Oversharing: participants may disclose more to an AI than intended. Remind them at the start what not to share, and explain how identifying details are removed. See our summary of research on oversharing with AI interviewers.
- Inappropriate follow-up questions: topic limits in the AI instructions, plus a pilot reviewed by a human.
- Re-identification: pseudonymize before analysis and avoid pasting raw transcripts into general-purpose AI tools. Our note on AI re-identification of anonymized interview data explains why removing names alone may not be enough.
7. Analysis and verification
Say which steps use AI (transcription, translation, first-pass coding, theme drafts) and how a human verifies each. Concrete commitments read better than general ones: "the lead researcher checks 20% of transcripts against audio," "a second coder reviews all AI-suggested codes," "every quote in the report is traced to its transcript." Reuse this text later in your methods section on LLM use.
AI Interview Ethics Application Checklist
Run this checklist before you submit:
- The information sheet states that an AI conducts the interview, in the first paragraph.
- The interview guide and the AI's topic limits are attached.
- Every processor is named in a data-flow table with location, retention and transfer safeguard.
- Vendor terms on model training and retention are quoted with the date checked.
- A DPIA is attached, or a screening note explains why none is needed.
- Stop rules, support resources and a named follow-up contact are described.
- The limits of confidentiality for disclosures of harm are stated in the consent text.
- Withdrawal and deletion steps are specific to the platform used.
- The human verification plan for transcripts, codes and quotes is concrete.
- Any funder or partner restriction on AI use has been checked.
Where Qualitati fits
Qualitati is a European qualitative research platform for universities and research companies, based in Paris. It runs AI-moderated interviews in text and voice in 10 languages, AI-moderated focus groups and conversational surveys, and supports analysis with ThemeLens thematic analysis and the QDA Workspace for coding.
For an ethics application, a few features matter. Every plan, including the free tier, can switch on an EU-only AI processing option in the profile settings, and Qualitati's privacy policy lists AI providers and international transfers, which gives you material for the data-flow table. Confirm which features your chosen processing mode covers, because not every feature is processed the same way. The pricing page publishes per-credit rates, and the free tier includes 30 monthly credits (they do not roll over) with no credit card, enough to pilot an interview guide before you submit. Qualitati supports your ethics review; it does not replace it, and your institution remains responsible for the approval decision.
Limitations and methodology concerns
Requirements differ by country, institution and discipline, and many of the sources above are campus-specific guidance rather than law. EDPB Guidelines 1/2026 were a consultation draft at the time of writing. An AI moderator also has methodological costs: it can miss non-verbal cues, probe less flexibly than an experienced interviewer, and may not suit highly sensitive topics or vulnerable groups at all. This article is general information, not legal advice. Have your data protection officer and ethics office review the final application.
Who this is for, and when not to use this approach
Who this is for: PhD students, research labs and research firms planning interview studies with an AI moderator and submitting to a REC or IRB.
When not to use this approach: studies with children, people in crisis, or topics where a participant may need immediate human support. In those cases a trained human interviewer, perhaps with an AI assistant in the background, is usually the safer design.
FAQ
Do I need ethics approval to use an AI interviewer?
Usually yes, if the study would need approval with a human interviewer. Several universities treat an AI tool that interacts with participants as part of the research procedure, so the AI's role must be described in the application.
Do I need a DPIA for AI-moderated interviews?
Not always. GDPR Article 35 requires one when processing is likely to be high risk, for example large-scale processing of special category data. Check with your data protection officer and record the decision either way.
What should the consent form say about the AI?
That an AI conducts the interview, what is recorded and transcribed, which organizations process the data and where, whether data trains AI models, and how to withdraw. Put the AI disclosure near the top.
Can participants withdraw after an AI interview?
They should be able to. State the deadline, for example before anonymization or analysis, and the exact deletion steps on the platform you use.
Should I name the AI model provider in the application?
Yes, where the vendor discloses it. Committees ask who processes the data, and naming the platform and its listed AI providers answers that directly.
Bottom line
A strong ethics application for AI-moderated interviews treats the AI as part of the method: describe its role, map every data flow, plan for distress and disclosure, and commit to human verification of AI outputs. Use the reviewer map and checklist above before you submit. When you are ready to pilot your guide, start free with 30 credits or view transparent pricing.