Should AI Use Be Described in a Research Ethics Application?

Should You Tell the Research Ethics Committee That You Plan to Use ChatGPT?

You are preparing a research ethics application for a qualitative study involving university instructors. The proposed methodology includes semi-structured interviews, transcription, and thematic analysis.

You also plan to use generative AI. It may help refine the interview guide, review transcripts for possible patterns, and organize preliminary analytical notes.

However, the ethics application does not contain a specific question asking whether ChatGPT or another AI tool will be used. You wonder whether this means disclosure is unnecessary.

Would mentioning AI create additional administrative requirements for an otherwise straightforward study? Or could omitting it prevent reviewers from understanding how participant information will actually be processed?

The answer depends less on whether AI is mentioned by name in the application form and more on whether its proposed role is relevant to the ethical evaluation of the research.

Describe AI Use When It Is Relevant to the Ethical Review

How to Describe Generative AI Use in a Research Ethics Application

Why Does AI Disclosure Matter in Research Ethics Review?

Research ethics review evaluates the proposed study, including relevant procedures, participant involvement, data handling, foreseeable risks, and safeguards.

When generative AI changes any of these elements, reviewers may need information about the system to evaluate the protocol adequately.

For example, a researcher might describe a study as involving interviews followed by thematic analysis. However, if complete identifiable transcripts will be transmitted to an external AI provider, that arrangement may be relevant to confidentiality, participant information, and data protection.

The phrase "thematic analysis" does not explain where the data will be processed or who may have access to them.

Disclosure therefore serves a substantive purpose. It helps reviewers understand what the researcher actually proposes to do rather than merely demonstrating that the researcher has acknowledged using AI.

Not Every AI-Assisted Task Requires the Same Level of Disclosure

AI can participate in research activities with very different ethical implications.

Using ChatGPT to suggest alternative titles from nonconfidential information is not equivalent to using an AI chatbot to interview participants or an external model to process sensitive records.

These are illustrative considerations rather than universal disclosure rules. The applicable institution determines the required information and review process.

Researchers should distinguish whether AI use requires ethics approval from what must be described in an application for a study already subject to review.

What Information Should the Application Include?

When AI use is relevant to the protocol, reviewers may need more than the name of the tool.

A useful description identifies the proposed activity, information involved, processing arrangements, researcher oversight, and safeguards.

The description should reflect the actual proposed workflow. Researchers should avoid listing safeguards that have not been implemented or assuming that a provider offers protections without verifying the applicable agreement.

Why Naming ChatGPT Is Not Enough

Simply stating "ChatGPT will be used for data analysis" may provide insufficient information for ethical evaluation.

Different AI products and service configurations can have different contractual, retention, access, and data-handling arrangements.

For example, the conditions governing a publicly accessible consumer service may differ from those of an institutionally managed enterprise environment.

Researchers should identify the actual service arrangement and verify its relevant conditions rather than assuming that the provider's general reputation establishes the security of every product.

Where the exact system has not yet been selected, the researcher should explain the proposed processing requirements and obtain the necessary institutional determination before using a service that handles participant information.

This is particularly relevant when assessing whether an AI provider becomes a new recipient of research data.

Describe the Data That AI Will Actually Receive

The ethical implications of AI-assisted analysis depend partly on the information submitted.

For example, asking AI to suggest possible coding categories from a general description of a study differs from uploading complete interview transcripts containing identifiable experiences.

Researchers should specify whether AI will receive raw recordings, full transcripts, selected excerpts, aggregated information, or entirely hypothetical examples.

They should also identify whether direct identifiers, contextual identifiers, or sensitive information may be present.

Removing names does not necessarily make qualitative data anonymous. A participant may remain identifiable through their professional role, department, or description of a distinctive event.

The application should therefore avoid describing pseudonymized information as fully anonymous without an appropriate assessment.

Explain Why AI Is Needed for the Proposed Task

Research ethics review may consider whether the proposed processing is proportionate to the research purpose and whether less intrusive alternatives are available.

For example, if the researcher only needs help developing a preliminary coding framework, submitting complete identifiable transcripts may not be necessary.

The researcher might instead use nonconfidential descriptions, limited appropriately protected excerpts, or an authorized local processing environment.

The justification should explain what AI contributes and why the selected workflow is appropriate.

Convenience and speed may be legitimate practical considerations, but they do not automatically outweigh additional confidentiality or participant risks.

This is connected to data minimization in AI-assisted analysis, particularly when personal information is involved.

Should Participant Information and Consent Materials Mention AI?

When AI materially affects how participant information is collected, processed, or used, researchers should consider whether the consent materials need to explain that involvement.

For example, participants may reasonably distinguish between interviews analyzed exclusively by a university research team and interviews submitted to an external AI provider.

Whether specific disclosure or additional consent is required depends on the applicable legal framework, institutional policies, and approved research arrangements.

Researchers should not assume that a general statement about data analysis automatically covers every AI-assisted processing activity.

At the same time, mentioning AI does not eliminate the need for appropriate authorization, security, and confidentiality safeguards.

The relationship between formal permission and participant expectations concerning AI use deserves separate consideration.

Explain How AI Outputs Will Be Verified

AI-generated interpretations can be inaccurate, incomplete, or insufficiently grounded in the source material.

Suppose a researcher plans to use AI to identify preliminary themes from interview transcripts. The application should not imply that the system will independently establish the meaning of participants' experiences.

Instead, the researcher may describe how AI suggestions will be reviewed against the original transcripts and evaluated according to the selected qualitative analytical approach.

Where AI produces classifications or other outputs that could affect participants, the application may need to explain how errors, bias, and inappropriate inferences will be managed.

Human oversight should be described as an actual procedure, not a general assurance that the researcher will remain responsible.

How Can AI Affect the Risk Assessment Section?

Introducing AI may change the nature, likelihood, or consequences of certain research risks.

For example, external processing may introduce confidentiality concerns, while automated participant interaction may create risks involving unexpected or inappropriate system responses.

AI-generated classifications may also create risks of misrepresentation or stigmatization, particularly when findings concern identifiable individuals or small communities.

Researchers should explain which risks are relevant to the proposed activity and what safeguards will address them.

They should not present every conceivable AI-related problem as though it necessarily applies to the study.

The assessment should be specific enough to explain how AI may change the ethical risk of the research.

What Does an Appropriate Ethics Application Description Look Like?

A useful description is specific, proportionate, and consistent with the actual research workflow.

Consider a hypothetical study in which the researcher plans to use an institutionally authorized AI system for preliminary organization of de-identified interview excerpts.

This wording is illustrative, not a substitute for an institution's required application format. It should be used only when the described safeguards and authorization actually exist.

A study using a different provider, identifiable information, or AI-mediated participant interaction would require a different description.

Can AI Help Draft the Ethics Application Itself?

Generative AI can help organize an application, improve clarity, and identify information that may be missing from a proposed description.

However, the researcher must verify every factual statement concerning the study, provider arrangements, consent procedures, and safeguards.

AI should not invent institutional approval, claim that data are anonymous without justification, or describe technical protections that have not been confirmed.

The resulting draft should be reviewed against the actual protocol and institutional instructions before submission.

What If the Application Form Does Not Ask About AI?

Ethics application forms vary across institutions, and some may not contain an AI-specific section.

However, researchers should still accurately describe material aspects of the proposed procedures in the relevant methodology, data management, confidentiality, or risk sections.

For example, if AI will process participant transcripts, omitting that arrangement merely because the form lacks an AI checkbox may leave the data processing description incomplete.

Where the appropriate placement is unclear, researchers should seek guidance from the reviewing office.

The objective is not to mention AI repeatedly throughout the application, but to ensure that reviewers have sufficient information to evaluate the proposed activity.

What If AI Use Changes After Approval?

Researchers may decide to introduce a new AI tool or expand an existing AI-assisted workflow after receiving ethics approval.

For example, a protocol may describe AI assistance with interview guide drafting, but the researcher later proposes using an external service to analyze identifiable transcripts.

The second activity raises different considerations and may not be covered by the original approval.

Researchers should follow institutional procedures for determining whether the change requires notification, amendment, or additional authorization before implementation.

Detailed decisions about introducing AI after ethics approval should be based on the actual change rather than the assumption that any earlier mention of AI authorizes every later use.

Disclosure Is Not the Same as Ethical Approval

Researchers sometimes treat transparency as though it resolves the ethical implications of AI use.

However, describing an external processing arrangement does not independently establish that the arrangement is permissible, secure, or consistent with participant consent.

Similarly, reporting AI use in a published article does not retroactively authorize a procedure that should have received institutional review before implementation.

Disclosure provides information for assessment. Approval or authorization depends on the applicable review process and whether the proposed conditions satisfy the relevant requirements.

Researchers should therefore distinguish accurately reporting AI use from obtaining permission to conduct the proposed activity.

Writing an Ethics Application That Accurately Describes AI-Assisted Interview Analysis

The revised application does not merely announce that AI will be used. It explains what the system will do, what information it will process, and how the proposed activity will be governed.

This allows reviewers to evaluate the actual procedure rather than infer its implications from the name of an AI tool.

Common Mistakes When Disclosing AI Use in Research Ethics Applications

Describe the AI Workflow Precisely Enough for Meaningful Ethical Review

When preparing an ethics application, determine whether AI affects the participant experience, research procedures, data processing, or foreseeable risks.

If it does, describe the activity in terms that allow reviewers to understand and evaluate the proposed arrangement. Avoid both vague statements and unnecessary technical detail unrelated to the ethical assessment.

Keep the application consistent with the consent materials, data management plan, and actual methodology.

A well-prepared disclosure should allow an ethics reviewer to understand what AI will do without having to guess how the technology fits into the research process.

Before Submitting an Ethics Application Involving Generative AI

Frequently Asked Questions About AI Disclosure in Research Ethics Applications

Disclose AI When It Matters to How the Research Will Be Conducted

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.