Would You Be Comfortable Explaining Your AI Use to an Editor, Reviewer, Participant, Funder, or Ethics Committee?

Could You Explain and Defend Your AI Use to Everyone With a Legitimate Interest in Your Research?

A researcher uses generative AI to prepare statistical code, classify interview responses, and improve manuscript wording. The resulting article is nearly ready for submission.

The researcher is comfortable telling the journal editor that AI assisted with writing. However, explaining the classification procedure to a methodological reviewer would be more difficult. The researcher has not preserved the original outputs or established how classification accuracy was evaluated.

The situation becomes more complicated when considering participants. The interview transcripts were processed through an external AI service, although the consent materials did not explicitly describe that arrangement.

Would the researcher be comfortable explaining these decisions to the journal, the participants, the research funder, or the institutional ethics committee?

The question is useful because discomfort may reveal an unresolved methodological or ethical concern. Yet comfort alone is not a reliable standard. A researcher can confidently explain an inappropriate procedure, just as a cautious researcher may feel uneasy about an entirely defensible one.

You Should Be Able to Give an Accurate, Evidence-Based Explanation of Consequential AI Use

How to Explain AI-Assisted Research to Different Stakeholders

Why the Ability to Explain AI Use Matters

Scientific research involves accountability to people and institutions with legitimate interests in how the work was conducted.

Editors assess compliance with publication policies and the integrity of submitted manuscripts. Reviewers evaluate methodological justification and evidentiary support. Participants have interests in how their information is used and how research activities may affect them.

Funders may establish conditions governing project methods, expenditure, confidentiality, and reporting. Ethics committees evaluate research procedures and participant protections within their applicable authority.

These responsibilities overlap, but they are not identical.

A useful explanation should address the questions that the particular stakeholder is entitled to ask, rather than provide the same generic declaration to everyone.

Why Comfort Is Not a Scientific Standard

Researchers sometimes use a personal transparency question as a practical test: Would I be willing to explain this procedure openly?

The question can reveal uncertainty about undocumented AI use, insufficient verification, or questionable information processing.

However, psychological comfort does not establish methodological validity.

A researcher may be comfortable relying on an AI-generated citation without realizing that the publication does not exist. Another researcher may feel uncomfortable describing an approved and carefully validated AI-assisted procedure because the technology remains unfamiliar.

The meaningful standard is whether the explanation can be supported by relevant evidence and satisfies applicable obligations.

What Should Every Consequential Explanation Establish?

Although different stakeholders require different information, researchers should be able to establish the basic role of AI in the study.

This includes identifying the task, the system used where known, the information processed, the scientific purpose, and the relationship between generated outputs and final research decisions.

Researchers should also explain relevant verification procedures, limitations, and authorization.

For example, saying that AI assisted with statistical analysis is less informative than explaining that AI generated preliminary Python code, which researchers inspected, corrected, and executed using a documented analytical procedure.

The explanation should accurately distinguish automated assistance from independent scientific judgment.

What Does a Journal Editor Need to Know?

Journal editors may need information about AI assistance to evaluate compliance with publication policies and the accuracy of the scholarly record.

Requirements vary across publishers and may change over time.

Some policies distinguish routine language correction from substantive generative AI assistance. Others require broader declarations.

Researchers should examine the target journal's current instructions and describe AI involvement according to those requirements.

Where AI materially contributed to methods or analysis, the manuscript may also need an adequate methodological description.

An AI disclosure statement does not replace the author's responsibility for accuracy, originality, appropriate attribution, and research integrity.

What Does a Peer Reviewer Need to Know?

A reviewer evaluating an AI-assisted method may ask how the procedure was selected, implemented, and validated.

For example, if AI classified open-ended survey responses, the reviewer may reasonably ask how the categories were defined, whether the model was evaluated against reference judgments, and how classification errors could affect the findings.

A statement that all outputs were manually checked may be insufficient without an explanation of the review procedure.

Researchers should be able to present methodological evidence rather than rely on the model's reputation or confidence.

Reviewers may also need to understand whether model updates, unavailable technical information, or data-processing restrictions limit reproducibility.

What Do Research Participants Need to Know?

Participants' information needs depend on the study and applicable consent and ethics requirements.

When AI directly interacts with participants or materially changes how their information is processed, researchers should determine what must be communicated before participation or before the new processing occurs.

For example, participants may need to understand that their interview recordings will be processed by an external transcription provider, where that information is material to informed participation.

Researchers should avoid misleading statements about whether an interaction involves a human or an automated system.

However, not every use of AI for manuscript editing creates a new participant notification obligation.

Participant-facing explanations should be accurate, comprehensible, and consistent with the approved research arrangements.

What Does a Research Funder Need to Know?

Funders may establish requirements concerning research methods, data management, reporting, intellectual property, and responsible use of project resources.

An AI-assisted activity may raise questions about whether the procedure falls within the approved project scope or requires additional authorization.

For example, a funded study may have contractual restrictions on sharing data with external service providers.

Researchers should distinguish ordinary scholarly disclosure from the specific obligations established by the funding agreement.

A journal's permission to use AI does not automatically override funder requirements.

Where the project involves a consequential methodological change, researchers should examine whether the funder requires notification or approval.

What Does a Research Ethics Committee Need to Know?

Ethics committees may need to understand how AI affects participant recruitment, consent, interactions, confidentiality, eligibility, or foreseeable risks.

For example, an AI chatbot introduced into an educational intervention may change what participants experience and the safeguards necessary for the study.

A committee may require information about the system's intended function, the data involved, human oversight, validation, and procedures for handling unexpected outputs.

The applicable review requirements depend on the institution, jurisdiction, and research design.

Researchers should not assume that an approved original protocol automatically covers every subsequent AI-related change.

How Do Stakeholder Expectations Differ?

The table identifies typical interests rather than universal reporting requirements. The authority and information needs of each stakeholder depend on the governing arrangements.

Why a Generic AI Disclosure May Be Inadequate

A statement such as generative AI was used during the research may be accurate but insufficiently informative.

It does not distinguish manuscript editing from automated measurement or explain whether confidential data were processed.

Researchers should describe the actual contribution in language appropriate to the audience.

For a journal, this may involve a concise declaration and a separate methodological explanation.

For an ethics committee, the relevant information may concern the processing environment, participant risks, and safeguards.

The objective is substantive transparency, not merely mentioning AI.

How Should Researchers Explain AI-Generated Statistical Code?

Researchers should distinguish generating code from conducting and validating an analysis.

For example, AI may produce an initial script while the research team selects the statistical model, examines assumptions, corrects the code, and independently verifies important calculations.

A defensible explanation should identify these respective contributions.

If the researcher cannot explain the model specification or why it is suitable for the data, the difficulty is methodological rather than merely communicative.

Additional expertise or verification may be necessary before the result can be defended.

How Should Researchers Explain AI-Assisted Qualitative Coding?

AI-assisted qualitative analysis may involve preliminary coding, classification, summarization, or interpretation.

The explanation should identify what the system generated and how researchers engaged with the original material.

For example, researchers may use AI to propose initial categories while developing final themes through reflexive analytical work.

The methodological explanation should follow the selected qualitative approach.

Researchers should not claim that automated agreement establishes validity under every qualitative methodology.

Nor should they represent AI-generated interpretations as independent human analysis.

What if the AI System Cannot Be Fully Explained?

Researchers may not have access to proprietary model weights, complete training information, or hidden system instructions.

This does not automatically make every use of the system scientifically unacceptable.

However, researchers should identify which details are unavailable and whether those limitations affect the task.

For example, an unavailable model snapshot may create a significant reproducibility concern when AI-generated classifications form the primary outcome variable.

The explanation should distinguish known information from assumptions.

Researchers should not invent technical details to make the procedure appear more transparent.

What if Stakeholders Ask for Confidential Prompts or Data?

Transparency obligations do not automatically authorize unrestricted disclosure of participant information or confidential research materials.

A reviewer may need methodological information without receiving identifiable interview transcripts.

Researchers may provide de-identified examples, controlled-access records, methodological summaries, or other permitted evidence where appropriate.

The relevant arrangements depend on confidentiality commitments, journal requirements, institutional policies, and applicable law.

Researchers should not resolve one transparency concern by creating a new unauthorized disclosure.

Can Researchers Explain AI Use Without Defending Every AI Suggestion?

Yes. Research workflows often involve exploratory suggestions that are subsequently rejected.

Not every abandoned wording suggestion or speculative idea requires detailed justification.

However, consequential AI recommendations that materially changed the study may require an explanation of the final decision.

For example, if AI suggested changing a statistical method, the researcher should be able to explain the independent methodological evidence supporting the change.

The objective is accountability for the completed research rather than exhaustive documentation of every exploratory interaction.

What if an Explanation Reveals a Research Problem?

Difficulty explaining a consequential procedure may reveal missing documentation, insufficient validation, or an unresolved authorization issue.

Researchers should investigate the underlying problem rather than simply improve the wording of the explanation.

For example, if a team cannot explain how AI-generated classifications were validated, drafting a more polished disclosure does not establish their reliability.

The appropriate response may involve additional analysis, methodological review, institutional consultation, or revision of the research claims.

Transparency is valuable partly because it can expose weaknesses that require substantive correction.

Does Willingness to Explain AI Use Establish Research Integrity?

No. Willingness to disclose is consistent with transparency but does not establish that the procedure was authorized, valid, or accurately reported.

A researcher may sincerely believe that a method is appropriate while misunderstanding its assumptions.

Conversely, a researcher may be reluctant to discuss AI because of uncertainty about changing institutional attitudes, even when the procedure was defensible.

Research integrity should be evaluated through actual conduct, evidence, applicable requirements, and the accuracy of the scholarly record.

Personal comfort is best used as an invitation to examine these matters, not as a substitute for doing so.

When Should Researchers Seek Clarification Before Proceeding?

Clarification may be necessary when disclosure policies are ambiguous, participant information processing is uncertain, or the proposed AI activity changes an approved research procedure.

Researchers should identify the authority responsible for the particular question.

Editorial offices can clarify journal submission requirements. Institutional privacy personnel can assess information-processing arrangements. Ethics committees can determine whether proposed changes require review under their procedures.

Methodological specialists can help evaluate scientific validity.

Seeking clarification from the appropriate authority is generally more useful than asking one stakeholder to resolve matters outside its responsibility.

What Does a Defensible Explanation Ultimately Demonstrate?

A defensible explanation identifies the actual AI contribution and connects it to appropriate authorization, verification, and human scientific judgment.

It should enable the relevant stakeholder to understand the procedure without relying solely on the researcher's confidence.

Researchers should be able to acknowledge limitations and explain what was done to manage them.

The objective is not to make every stakeholder approve of AI use in principle.

It is to demonstrate that the particular use was appropriate, accurately reported, and supported by evidence sufficient for its scientific role.

Explaining AI-Assisted Interview Analysis to Different Research Stakeholders

Suppose the team discovers that the AI coding procedure was adequately validated but that the approved protocol did not clearly authorize the external processing service.

The correct response is not merely to disclose the provider in the manuscript. The researchers should seek the appropriate institutional determination and address any identified noncompliance.

The example demonstrates why different stakeholders may require different evidence about the same AI-assisted activity.

Misconceptions About Explaining AI Use to Research Stakeholders

Prepare Explanations That Match the Stakeholder's Legitimate Questions

Before completing an AI-assisted study, identify who may reasonably need information about the AI contribution and what each stakeholder is entitled to evaluate.

Researchers should connect AI disclosure requirements with the evidence needed to defend AI-assisted methods during peer review.

Researchers should also ensure that consequential AI interactions are documented and that the analytical process can be reconstructed where feasible.

The strongest explanation is not necessarily the longest. It is the one that accurately identifies what AI did and provides the evidence needed to evaluate its contribution.

Before Explaining AI Use to Research Stakeholders

Frequently Asked Questions About Explaining AI Use in Research

Be Prepared to Explain AI Use With Evidence, Not Merely Confidence

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.