Can AI Use Conflict With Participant Expectations Even When It Is Technically Permitted?

Can AI Use Be Ethically Questionable Even When the Researcher Is Allowed to Use It?

You are conducting interviews with university instructors about their experiences with institutional policies concerning generative AI. Participants have agreed to be interviewed, and the consent form explains that their responses will be analyzed for research purposes.

Your university provides access to an approved AI platform, and its contractual arrangements permit certain categories of research data processing. You decide to use the system to help organize interview transcripts and identify preliminary analytical patterns.

From an institutional and technical perspective, the arrangement may be permitted. However, imagine that a participant later discovers that their interview was processed by an AI system.

The participant responds, "I agreed to share my experiences with your research team. I did not realize that an AI system would analyze what I said."

The researcher may have complied with the available technical permissions, but the participant's reaction raises a different question. Does having permission to use AI necessarily mean that its use is consistent with what participants understood and reasonably expected?

Yes, Technical Permission and Participant Expectations Can Differ

Why Participant Expectations Matter in AI-Assisted Research

What Are Participant Expectations in Research?

Participant expectations concern what individuals understand or reasonably anticipate about their involvement in a research study.

These expectations may include who will conduct the research, what information will be collected, who will have access to it, how it will be analyzed, and how the findings will be reported.

They are influenced by consent materials, recruitment explanations, institutional relationships, previous research experiences, and the circumstances under which information is shared.

For example, a university instructor who agrees to a confidential interview may expect the research team to examine the transcript using established qualitative analytical procedures.

The participant may not necessarily expect the transcript to be submitted to an external generative AI service, even when that service operates under contractual confidentiality protections.

However, researchers should not assume that every participant opposes AI processing. Some may welcome its use, while others may have reservations or require additional explanation.

The ethical task is to consider reasonable expectations without treating any hypothetical reaction as representative of the entire participant population.

Technical Permission Is Not the Same as Ethical Acceptability

Researchers may obtain permission to use an AI system through an institutional subscription, data processing agreement, or approved research protocol.

These permissions establish important conditions, but they do not necessarily address every aspect of the participant's understanding of the research.

For example, an institution may authorize an enterprise AI platform to process certain research records. Nevertheless, a participant may have understood that their interview would be examined only by human researchers.

That difference does not automatically establish a consent violation or prove that the processing is unlawful. It does, however, provide a reason to examine whether the participant information was sufficiently clear and whether the processing respects the research relationship.

Researchers should distinguish legal compliance, institutional authorization, and ethical justification rather than treating them as interchangeable.

Why Contextual Integrity Is Relevant

Helen Nissenbaum's theory of contextual integrity provides a useful framework for examining whether information flows are appropriate within particular social contexts.

The framework considers how information moves among actors according to contextual norms, including the nature of the information, the parties involved, and the conditions governing its transmission.

Applied to research, this perspective suggests that the ethical significance of data processing depends partly on the circumstances in which participants provided their information.

For example, a participant may willingly discuss sensitive professional experiences with a researcher under an expectation of confidentiality. The same participant may have different expectations about submitting those experiences to an AI provider.

The information itself has not necessarily changed, but the actors and conditions involved in processing it may have.

Contextual integrity does not establish a universal prohibition against AI processing. Rather, it provides a way to examine whether the proposed information flow is appropriate within the research context.

Nissenbaum (2004) discusses privacy as contextual integrity, emphasizing that privacy cannot be understood solely through whether information is secret or disclosed.

What Participant Expectations Might AI Affect?

AI may change different aspects of how participants understand the research process.

These differences may or may not create an ethical problem in a particular study. Their significance depends on what participants were told, the research purpose, applicable permissions, and potential consequences.

Can Broad Consent Language Cover AI Processing?

Consent forms sometimes state that participant information will be used for research analysis without specifying every analytical technique.

Whether such language permits a particular AI-assisted activity depends on the governing ethics and legal framework, the scope of consent, and the approved protocol.

Researchers should avoid assuming that broad language automatically authorizes any future processing method.

However, they should also avoid claiming that every software change necessarily requires renewed consent.

The more important question is whether the proposed AI use materially changes the information flow, risks, or purposes participants were informed about.

Where the answer is uncertain, researchers should seek guidance from the appropriate institutional authority.

Formal research participation consent should also be distinguished from the legal basis for personal data processing, which may not be identical under every data-protection framework.

Why Participants May Care Who or What Analyzes Their Responses

Participants may regard the identity and role of the analyst as relevant to their willingness to share information.

For example, an instructor discussing a sensitive disagreement with university administrators may be comfortable speaking to a researcher who has explained the confidentiality procedures.

The same instructor may have reservations about an automated system processing the transcript, particularly if the provider's role or data retention arrangements are unclear.

Other participants may be more concerned about how AI interprets their statements than about where the data are processed.

They may worry that a model will classify their experiences inaccurately or infer characteristics they did not intend to disclose.

Researchers should not dismiss these concerns simply because the AI system is technically capable of performing the task.

At the same time, participant expectations should be investigated or considered contextually rather than invented by the researcher or AI model.

AI Can Change the Meaning of Confidentiality Assurances

Researchers often assure participants that their information will be kept confidential and accessed only under specified conditions.

When AI is introduced, the researcher should determine whether those assurances remain accurate.

For example, a statement that "only the research team will access your interview transcripts" may be difficult to reconcile with a proposed arrangement in which an external provider processes the full transcripts, depending on the wording and technical conditions.

However, external processing does not automatically mean that information becomes public or that confidentiality has been breached.

An authorized service may operate under contractual restrictions and technical safeguards.

The ethical concern is whether the researcher's explanation accurately reflects the processing arrangement and whether participants have been given information appropriate to the decision they are making.

Researchers should examine the AI provider's role as a possible recipient or processor of research data before making confidentiality assurances.

Does AI Use Need to Be Explained in Technical Detail?

Participant information should be sufficiently clear to support an informed decision, but it need not reproduce the complete technical architecture of an AI system.

For example, participants may benefit from knowing whether their interview responses will be processed by an external AI service, what the system will be used for, and what safeguards apply.

They may not need a detailed explanation of model parameters or computational architecture when those details do not materially affect their participation.

The appropriate level of detail depends on the proposed activity and the information relevant to participants' decisions.

Researchers should avoid both extremes: concealing material processing arrangements behind vague terminology and overwhelming participants with technical information that obscures the practical implications.

Clarity is particularly important when participants have different levels of familiarity with generative AI.

How Should Researchers Describe AI Use to Participants?

A useful explanation identifies the role of AI, the information it will process, and the safeguards relevant to the participant.

For example, a study may propose using an authorized AI system to help organize de-identified interview excerpts while retaining researcher responsibility for interpretation.

This wording is illustrative and should only be used when the described authorization and safeguards actually exist.

Researchers should follow institutional guidance concerning whether additional information, participant options, or specific consent provisions are required.

Can AI Use Conflict With Expectations Without Violating the Consent Form?

Potentially. Formal consent documentation may satisfy applicable requirements while leaving participants with assumptions that differ from the researcher's intended practices.

For example, a consent form may broadly authorize analysis using approved research technologies. Nevertheless, participants may not realize that this includes generative AI.

This does not automatically invalidate the consent or establish misconduct.

However, it may indicate that additional explanation would support transparency and respect for participants.

The ethical significance depends on whether the difference concerns material information, the potential consequences, and the requirements governing the study.

Researchers should not treat the absence of an explicit prohibition as sufficient evidence that participants would reasonably expect every permitted activity.

Should Researchers Ask Participants Whether They Accept AI Analysis?

In some studies, asking participants about AI processing may help clarify expectations and inform consent procedures.

For example, a research team might consult representatives of the intended participant population when developing information materials for a study involving sensitive interviews.

Such consultation may reveal terminology that participants find confusing or information they consider important.

However, consultation with a small number of individuals does not necessarily establish the preferences of every future participant.

Nor should researchers assume that an informal expression of comfort with AI substitutes for the consent and authorization requirements governing the study.

Where participant choice about AI processing is offered, the research team should ensure that the options are genuine and operationally feasible.

For example, offering an opt-out from AI processing would be misleading if the research workflow cannot actually accommodate it.

How Can AI Generate Inferences That Exceed Participant Expectations?

Participants may agree to provide information about a particular topic without anticipating that researchers will infer additional personal characteristics.

For example, instructors may consent to interviews about their decisions concerning generative AI in teaching.

An AI system might subsequently generate speculative interpretations about participants' professional competence, emotional states, or willingness to comply with institutional rules.

These inferences may exceed the intended research purpose and may not be supported by the data.

The ethical concern is not resolved simply because the inferences were derived from information participants voluntarily provided.

Researchers should consider whether AI-generated sensitive inferences are necessary, authorized, and consistent with what participants were told.

Where the inferences are unnecessary or unjustified, they should not be generated or used merely because the model can produce them.

Can Participant Expectations Differ Across Research Contexts?

Yes. Expectations may differ according to the research topic, participant population, institutional setting, and relationships involved.

For example, participants in a study explicitly examining AI-mediated communication may reasonably anticipate more extensive AI involvement than participants in a confidential interview study unrelated to AI.

Similarly, individuals discussing publicly available professional practices may have different expectations from those describing sensitive workplace conflicts.

Researchers should not assume that one disclosure statement will be equally appropriate for every study.

Contextual evaluation is necessary because participants' understanding of acceptable information flows may depend on why they provided the information and who they expected to use it.

However, researchers should avoid stereotyping participant groups or predicting their preferences solely from demographic characteristics.

How Should You Prompt AI to Review Participant Expectation Concerns?

AI can help identify potential differences between the described research procedures and what participants may reasonably understand.

However, the model cannot establish actual participant expectations without evidence from the relevant population.

This prompt can support a preliminary review of the materials, but any substantive concerns should be evaluated through the appropriate institutional process.

Researchers should not upload identifiable participant information to an unauthorized AI service while seeking advice about whether such processing is appropriate.

What If Participants Object to AI Use?

If participants express concerns about AI processing, researchers should respond according to the study's consent arrangements, institutional requirements, and the stage of the research.

For example, a participant may ask whether their interview will be processed by an external AI service or whether they can decline that form of analysis.

The researcher should provide accurate information rather than make assurances that have not been verified.

Whether the participant can withdraw particular data, request an alternative processing arrangement, or exercise other rights depends on the applicable consent conditions and legal framework.

Researchers should not promise that information can always be withdrawn after analysis or that every provider can immediately delete all previously processed data.

Where the concern reveals a possible mismatch between the approved protocol and actual processing, the researcher should consult the appropriate institutional authority.

What If AI Use Was Introduced After Participants Consented?

Introducing AI after data collection may create new questions about participant expectations because the original information materials may not have described the proposed processing.

For example, a researcher may have obtained consent for interviews analyzed by the research team but later decide to use an external AI service.

The researcher should review whether the proposed change is covered by the original protocol and consent arrangements before implementing it.

Depending on institutional requirements, the change may require notification, amendment, additional safeguards, or further participant information.

Researchers should follow the applicable procedures for introducing AI after ethics approval rather than assuming that an unchanged research question makes the new processing arrangement ethically equivalent.

Why Trust Matters Beyond Formal Compliance

Research depends partly on participants' willingness to provide information under the conditions explained to them.

When participants discover an unexpected use of their information, they may question whether the research team accurately represented the study, even if the processing arrangement was technically permitted.

Such concerns can affect confidence in the research relationship and, potentially, willingness to participate in future studies.

However, researchers should not assume that AI involvement necessarily reduces trust. Participants may evaluate AI use differently depending on the purpose, transparency, safeguards, and their own experiences.

The ethical responsibility is to provide appropriate information and avoid creating a misleading impression about how the research will be conducted.

Respect for participants should not be reduced to satisfying the minimum wording of a consent document when material aspects of the processing may remain unclear.

When Institutionally Permitted AI Analysis Surprises a Research Participant

The hypothetical concern is not that an authorized AI provider necessarily mishandles the data. It is that participants may not have understood the processing arrangement in the same way as the researcher.

Addressing that difference through appropriate information and institutional review may support ethical transparency without assuming that every participant has identical expectations about AI.

Common Misconceptions About AI Use and Participant Expectations

Consider What Participants Understood, Not Only What the System Allows

When planning AI-assisted research, examine the proposed processing from the participant's perspective as well as the institution's.

Identify what participants were told about data access, analysis, confidentiality, and the involvement of external services. Then compare those explanations with what the AI-assisted workflow will actually involve.

Document any material differences identified between the participant information and the proposed processing arrangement, together with the institutional determination and resulting safeguards.

The objective is not to obtain participant approval for every minor software operation. It is to avoid introducing consequential processing arrangements that participants were not adequately informed about.

Before Using AI With Information Provided by Research Participants

Frequently Asked Questions About AI and Research Participant Expectations

Permission to Use AI Does Not Automatically Mean Participants Expected It

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.