03 · What You Need to Know
How AI Can Help Review Sensitive Questions in Qualitative Research
What Makes an Interview Question Sensitive?
An interview question may be sensitive when answering it could involve emotional discomfort, disclosure of private information, social stigma, professional consequences, or other risks to the participant or people mentioned in the response.
Sensitivity is not determined solely by the subject matter. The same question may have different implications depending on who is asked, where the interview occurs, and how the information will be recorded or reported.
For example, asking an instructor about experiences with institutional AI policies may appear routine. However, if the participant works in a small department and has previously disagreed with administrators, describing a particular incident could make them identifiable.
Lee (1993), in Doing Research on Sensitive Topics, examines methodological and ethical challenges associated with research involving potentially threatening, private, or socially consequential information.
AI can help identify possible sources of sensitivity, but it cannot determine the actual consequences of disclosure without sufficient contextual information.
Different Questions Can Create Different Types of Risk
Researchers should examine the specific concern associated with a question rather than treating sensitivity as one undifferentiated category.
| Potential Concern |
Illustrative Question |
Why It May Be Sensitive |
| Emotional distress |
Can you describe an experience when your professional judgment was publicly criticized? |
May involve recalling a distressing incident. |
| Professional consequences |
Have you ever used AI in ways that violated institutional policy? |
May invite disclosure of conduct with employment implications. |
| Privacy |
Can you describe personal circumstances that affected your teaching performance? |
May invite disclosure of private information unrelated to the research need. |
| Third-party identification |
Which colleagues have misused generative AI? |
May identify people who have not consented to participate. |
| Institutional conflict |
Can you describe disagreements with your department about AI use? |
May reveal sensitive workplace relationships. |
| Power relationships |
How did your supervisor respond when you questioned the AI policy? |
May create concerns about identification or retaliation. |
These examples illustrate possible concerns rather than establishing that the questions are always inappropriate. Some sensitive questions may be essential to the research purpose and ethically justifiable when appropriate safeguards are in place.
Distinguish Question Sensitivity From Participant Vulnerability
Researchers should avoid assuming that particular participant groups will necessarily find certain questions distressing or difficult.
For example, a question about professional conflict may be uncomfortable for one instructor but relatively straightforward for another. Differences may reflect personal experiences, employment conditions, institutional relationships, or willingness to discuss the topic.
Similarly, membership in a particular demographic or occupational group does not automatically determine how an individual will respond.
Potential question sensitivity
Characteristics of the topic or wording that may create risks or discomfort in particular circumstances.
Actual participant response
How an individual experiences and responds to the question during the research interaction.
AI can assist with the first assessment. It cannot establish the second before the interaction occurs.
Researchers should therefore avoid treating AI-generated labels such as "high sensitivity" as predictions of individual emotional reactions.
Can AI Identify Hidden Disclosure Risks?
Some questions appear neutral but may invite information that identifies participants or third parties.
Consider the question: "Can you describe a situation when your department handled an AI-related academic integrity incident?"
In a large institution, a general account might be difficult to attribute to a particular person. In a small department where only one incident occurred, the same description could identify the people involved even if names are omitted.
AI can flag the possibility of deductive disclosure, where individuals may be identified through combinations of contextual details rather than direct identifiers.
However, assessing the likelihood of identification requires knowledge of the actual institution, participant population, and planned reporting practices.
Researchers should consider whether particular details are necessary for answering the research questions and how identifiable information will be handled.
How Should You Prompt AI to Review Interview Question Sensitivity?
A useful prompt should provide the research purpose, participant characteristics, interview setting, and relevant ethical considerations. It should ask AI to identify possible risks without presenting them as confirmed participant reactions.
Illustrative AI Prompt
Screening a Qualitative Interview Guide for Sensitive Questions
"Review the following qualitative interview guide for questions that may involve emotional discomfort, private information, professional consequences, identifiable third parties, or sensitive institutional relationships. The participants are university instructors discussing experiences with generative AI in teaching. For each potentially sensitive question, identify the possible source of risk, explain what contextual information is needed to assess it, and suggest alternative wording or procedures where appropriate. Distinguish sensitivity from leading language and avoid assuming that all participants will react similarly. Do not assign definitive risk levels or claim that the guide is ethically approved. Identify concerns requiring researcher judgment or institutional ethics review."
This prompt encourages AI to identify possible mechanisms of risk rather than simply classify questions as safe or unsafe.
The researcher should verify the suggestions against the approved research protocol and the circumstances of the intended participants.
Can Changing the Wording Reduce Sensitivity?
Wording can influence whether a question appears accusatory, judgmental, or unnecessarily intrusive.
For example, "Have you ever violated your university's AI policy?" directly invites disclosure of potentially sanctionable conduct.
If the study concerns how instructors interpret institutional expectations rather than individual violations, a less intrusive question might ask: "Can you describe situations in which institutional expectations about AI use were difficult to interpret or apply?"
However, this revision changes the information being requested. It no longer directly investigates personal policy violations.
Researchers should not assume that softer wording is automatically equivalent to the original question. The revision must still address the research purpose.
AI can help identify leading interview questions, but sensitivity and leading language are different concerns that may require different responses.
Should Sensitive Questions Always Be Asked at the End?
Placing sensitive questions after participants have become familiar with the interview may sometimes support a more comfortable conversation.
However, there is no universal rule that sensitive questions must appear at the end.
If the central research topic involves a difficult experience, delaying it until the final minutes may leave insufficient time for careful exploration, clarification, or an appropriate closing.
Conversely, introducing highly personal questions immediately may be inappropriate when participants have not yet had an opportunity to understand the conversational context.
The appropriate sequence depends on the research purpose, participant population, consent procedures, and potential consequences of disclosure.
AI can help sequence interview questions, but ethical suitability cannot be determined through question order alone.
What Does Trauma-Informed Interviewing Contribute?
When research involves potentially distressing experiences, trauma-informed principles may help researchers consider participant safety, choice, trust, and control over disclosure.
These principles should not be interpreted as assuming that every participant has experienced trauma or that all difficult questions will cause harm.
Rather, they encourage researchers to avoid unnecessary coercion, communicate the purpose of questioning clearly, and respect participants' decisions about what they wish to discuss.
For example, participants may be reminded that they can decline a question or request a pause, consistent with the approved consent process.
Researchers should also prepare appropriate procedures for responding if a participant becomes distressed.
AI can suggest such considerations, but the final procedures must reflect the study's ethical approval, institutional requirements, and available support arrangements.
Can AI Predict Which Participants Will Become Distressed?
No. AI cannot reliably determine how a particular individual will respond emotionally to an interview question.
A model may infer possible reactions from general descriptions of the topic or participant population, but those inferences are not observations of the individual.
For example, an instructor may discuss a difficult institutional disagreement without distress, while another may find an apparently routine question about professional evaluation uncomfortable.
Researchers should not use AI-generated predictions to label participants as emotionally vulnerable or exclude them without an appropriate ethical and methodological basis.
Instead, interview procedures should provide reasonable safeguards and allow participants to communicate their preferences during the interaction.
When Should a Sensitive Question Be Removed?
A potentially sensitive question should not automatically be removed merely because it involves a difficult topic.
Some research questions legitimately require exploring experiences involving conflict, stigma, institutional authority, or personal hardship.
The more relevant consideration is whether the information is necessary for the research purpose and whether the risks are justified and appropriately managed.
For example, a study examining instructors' experiences of institutional AI enforcement may need to ask about difficult policy-related encounters. However, requesting names of colleagues accused of misconduct may be unnecessary.
Researchers should distinguish essential sensitive content from avoidable requests for identifiable or irrelevant information.
AI can help generate alternatives, but the decision requires ethical and substantive judgment.
What About Sensitive Questions in Focus Groups?
Sensitivity can be especially consequential in focus groups because disclosures occur in the presence of other participants.
Even when the researcher protects recordings and transcripts, participants may repeat information outside the discussion. Confidentiality therefore cannot be guaranteed in the same way as information controlled solely by the research team.
Questions involving personal misconduct, workplace conflict, or identifiable third parties may be less appropriate for group discussion than for confidential individual interviews.
Researchers should consider whether the topic can be discussed at a general level or whether individual interviews would provide a more suitable setting.
AI can help develop focus group discussion guides, but group composition and confidentiality limitations require particular attention when sensitive topics are involved.
Can Pilot Interviews Reveal Sensitivity Problems?
Pilot interviewing can help researchers examine whether questions are understandable, whether transitions feel abrupt, and whether the wording creates unnecessary discomfort.
However, a pilot interview cannot guarantee that future participants will react similarly. Sensitivity may vary across individuals and contexts.
Researchers should use pilot feedback to refine the guide while maintaining procedures that allow participants to decline questions or pause the interview when appropriate.
AI simulations may help prepare hypothetical scenarios, but they do not establish actual participant reactions.
The need for pilot testing an interview guide should therefore be considered alongside, rather than replaced by, AI-assisted risk screening.
Protect Sensitive Research Materials When Using AI
Researchers should consider confidentiality before submitting interview guides, participant descriptions, or excerpts from actual interviews to external AI services.
A draft guide may contain information about restricted institutional investigations or sensitive research populations. Actual participant responses may include identifying details, allegations, or personal information.
Such materials should not be uploaded without appropriate authorization and safeguards consistent with the study's consent arrangements and applicable data-protection requirements.
Where possible, researchers can ask AI to review generalized or hypothetical questions rather than sharing identifiable participant accounts.
If AI materially contributes to ethical screening or question development, its role should be documented according to applicable institutional and publication requirements.
