The adoption of AI moderation in qualitative research has been fast, and the ethical frameworks around it are still catching up. This creates genuine risk for research teams and their clients. Not just regulatory risk, but the risk of conducting research that respondents would not have consented to if they had fully understood what was happening.
This article is not an attempt to discourage AI-moderated research. Used responsibly, it produces better data faster at lower cost. But “responsibly” requires some specificity, and the specificity is worth working through.
What Makes AI-Conducted Research Ethically Distinct
Traditional qualitative research ethics are well established. Participants are informed about the purpose of the research, give their consent, can withdraw at any time, and have reasonable expectations about how their data will be used. These norms developed over decades in the context of human-to-human research interactions.
AI moderation introduces three new dimensions that existing ethics frameworks did not anticipate.
The first is the nature of the interaction. A respondent who believes they are talking to a person and discovers they are talking to an AI will feel deceived, even if they consented to “research participation.” The nature of who, or what, is asking the questions matters to most people. Disclosure is not optional.
The second is the nature of the data. AI-moderated interviews generate verbatim transcripts automatically, which are often processed by a large language model to generate summaries and themes. The respondent’s words travel further and through more processing layers than in a traditional interview. They need to know this.
The third is the scale. AI moderation makes it practical to interview hundreds of people quickly. The scale that makes it powerful also makes responsible data handling more important. A breach of 200 respondent transcripts is a different order of problem than a breach of 15.
Informed Consent in the AI Moderation Era
Informed consent for AI-moderated research should include four elements:
- Disclosure that the interview will be conducted by an AI system, not a human researcher
- Explanation of what data will be collected (verbatim transcript, any voice notes if applicable)
- Explanation of how the data will be stored, who will have access to it, and when it will be deleted
- Clear opt-out mechanism: the respondent can stop the interview at any time without consequence
The consent process should be simple, written in plain language, and presented before the interview begins. Burying disclosure in a terms and conditions page that 95% of respondents will not read is not informed consent in any meaningful sense.
Merren’s standard consent flow presents a plain-language disclosure message at the start of every interview. It requires an affirmative response before proceeding and gives respondents a clear way to end the session at any point.
Data Storage, Retention, and Participant Rights
The two regulatory frameworks most relevant to research conducted in or for India are India’s Digital Personal Data Protection Act (DPDP) and, for any data involving EU residents, the GDPR.
Key requirements under India’s DPDP Act (2023)
- Data fiduciaries (the research agency or client) must obtain clear consent before processing personal data
- Respondents have the right to access their data, correct inaccuracies and request deletion
- Data must be deleted once the purpose for which it was collected is served
- Any cross-border transfer of data requires compliance with central government rules (still being developed as of early 2026)
Key requirements under GDPR (for EU-based respondents)
- Consent must be freely given, specific, informed, and unambiguous
- Data subjects have the right to erasure (“the right to be forgotten”)
- Data minimisation: collect only what is necessary for the stated purpose
- Data processing agreements are required when using third-party AI platforms
For most research teams, the practical implication is straightforward: do not retain identifiable respondent data beyond the analysis period. Anonymise or delete transcripts once your thematic analysis is complete. Document your data handling process so you can demonstrate compliance if challenged.
Bias in AI Moderation
Data privacy is the compliance dimension of AI research ethics. Bias is the methodological dimension, and it receives less attention than it deserves.
AI moderation systems can introduce bias in three ways. The first is in the language of the guide: if the discussion guide contains leading questions or culturally loaded assumptions, the AI will faithfully execute those biases at scale across every respondent. There is no human moderator to intuitively compensate.
The second is in the AI’s probing behaviour. If the model that powers the AI moderator was trained predominantly on English-language Western data, its sense of what constitutes a complete answer will not match the communication norms in different cultural contexts. This can cause the AI to probe on things that do not need probing and miss things that do.
The third is in the automated analysis. When an AI summarises 60 transcripts into five themes, it is making choices about what counts as significant. Those choices reflect the model’s training data and architecture. A researcher who treats AI-generated theme maps as objective outputs rather than as one analytical lens among several is misunderstanding the tool.
Best Practices for Ethical AI Research
A checklist for any team deploying AI-moderated research:
- Disclose AI involvement before the interview begins, in plain language
- Obtain affirmative consent before proceeding
- Store transcripts in encrypted systems with access controls
- Set a data retention period and enforce it
- Anonymise data before sharing with clients unless the respondent has specifically consented to identified sharing
- Review AI-generated analysis against raw transcripts before presenting to stakeholders
- Test your discussion guide for leading questions before deploying at scale
- Ensure your AI platform has documented data processing agreements
For teams using AI moderation for the first time, the question of synthetic data is often raised at this stage. See What is Synthetic Data in Research? for a clear explanation of the distinction between AI-moderated research and synthetic data research, which are frequently confused.
How Merren Approaches Research Ethics
Merren is built on the principle that faster research should not mean less ethical research. Maya AI discloses its AI nature at the start of every interview. Consent is captured as an affirmative response before the interview begins. Transcripts are stored in encrypted systems and access is restricted to the client team and authorised Merren personnel.
We support data deletion requests and operate data retention policies that comply with India’s DPDP Act. For clients running research involving EU-based respondents, we provide data processing agreements that satisfy GDPR requirements.
Our bias mitigation approach includes guide review before deployment. Cross-cultural tone testing for Hindi-language interviews and a policy of presenting automated analysis alongside, not instead of, analyst-reviewed findings.
Research ethics is not a compliance checkbox. It is the foundation of trust between researchers, respondents, and the clients who rely on the data. If respondents do not trust the research process, the data they produce is less honest and therefore less valuable. Ethical research is better research.