Which Types of AI Use in Research Usually Require Disclosure?

Which Uses of AI Are Significant Enough to Declare in a Research Paper?

A researcher uses generative AI throughout a project. ChatGPT suggests possible research questions, an AI-powered search tool helps identify publications, another system proposes codes for interview transcripts, and a language model improves the manuscript's grammar.

Should all these activities be disclosed? Is using AI to correct a sentence equivalent to using it to analyze research data? What if an AI-generated suggestion influences the study but no generated text appears in the manuscript?

These distinctions matter because disclosure requirements are not determined simply by whether a researcher opened an AI application. The purpose of the assistance, its effect on the research, and the applicable publishing policy all influence what must be reported.

Researchers therefore need a practical way to distinguish routine assistance from AI involvement that materially affects the scholarly work.

Substantive Writing and Methodological AI Assistance Commonly Require Disclosure

The most important distinction is whether AI merely assisted with presentation or materially influenced the intellectual or methodological substance of the research. This distinction helps determine what information readers may need, although the governing policy ultimately determines the formal disclosure requirement.

How Disclosure Requirements Differ Across Research Activities

Why Is There No Universal List of AI Uses That Must Be Disclosed?

Scholarly publishers do not all apply identical AI-use policies.

The International Committee of Medical Journal Editors (ICMJE) recommends disclosure of AI-assisted technologies used in producing submitted work and distinguishes writing assistance from methodological applications.

The Committee on Publication Ethics (COPE) emphasizes transparency about AI use in manuscript writing, image production, and research activities.

Springer Nature distinguishes certain routine AI-assisted copyediting activities from substantive generative AI use, while Elsevier provides its own requirements concerning generative AI in manuscript preparation.

These differences mean that an activity requiring a formal statement under one publisher's policy may receive different treatment elsewhere.

Researchers should also consider institutional, funding, ethics, and contractual obligations that may apply independently of journal instructions.

The relevant question is therefore not only when researchers should disclose AI use, but which requirements govern the particular activity.

How Can AI-Assisted Research Activities Be Classified?

A useful distinction is between assistance with manuscript presentation, substantive content development, and research methodology.

These categories are analytical aids rather than universally recognized regulatory classifications.

These reporting locations are illustrative. A journal may require a different section or a separate declaration even when the activity is correctly classified.

Does AI-Assisted Manuscript Drafting Usually Require Disclosure?

Substantive generative AI drafting is among the activities most clearly addressed by publisher disclosure policies.

For example, a researcher may use ChatGPT to generate an introduction, prepare a preliminary literature review, or draft parts of the discussion.

Such assistance differs from correcting typographical errors because the AI system contributes new wording and may introduce claims, interpretations, or argumentative structure.

The ICMJE recommends reporting AI writing assistance in the acknowledgment section. Other publishers may require a designated generative AI statement or another reporting location.

Researchers should also verify generated content and ensure that its use complies with the journal's policy.

Disclosure does not automatically authorize every form of AI-generated drafting.

Does AI-Assisted Grammar Correction Usually Require Disclosure?

Not necessarily. Routine AI-assisted grammar correction may be exempt under some publisher policies.

Springer Nature, for example, distinguishes AI-assisted copyediting from more substantive generative AI activities under its stated conditions.

However, the boundary between copyediting and drafting can be difficult to establish.

A tool asked to correct grammar may substantially rewrite sentences, alter disciplinary terminology, or introduce new interpretations.

Researchers should therefore evaluate the actual output rather than relying solely on the prompt used.

The question of whether AI-assisted grammar correction requires disclosure depends on both the nature of the changes and the applicable policy.

Does AI-Assisted Translation Require Disclosure?

AI-assisted translation may require disclosure, particularly when it involves substantive manuscript translation or affects research data.

Translation is not always a mechanical substitution of words. It can involve decisions about meaning, disciplinary terminology, and cultural context.

For example, translating an interview transcript with AI may influence how participant responses are interpreted during analysis.

Such methodological use may warrant reporting even when the translated text does not appear directly in the final manuscript.

Translation of manuscript prose may be governed by language-editing or generative AI policies, depending on the publisher.

Researchers should verify translated content and consult the requirements for disclosing AI-assisted translation rather than assuming that all translation tools are treated identically.

Does Brainstorming With Generative AI Need to Be Disclosed?

Brainstorming occupies a less clearly defined area in many AI-use policies.

Researchers may use AI to suggest possible research questions, compare theoretical perspectives, or generate preliminary project ideas.

Some such interactions may have little lasting influence on the research. Others may substantially shape the study's conceptual framework or methodological direction.

The absence of copied AI-generated text does not necessarily mean the tool made no substantive contribution.

However, not every exploratory interaction automatically requires formal disclosure.

Researchers should assess the significance of the assistance and follow the applicable policy.

Where an AI-generated suggestion materially influences research design, maintaining a record of that contribution may support later transparency.

Does AI-Assisted Literature Searching Require Disclosure?

AI-assisted literature searching can be methodologically significant, especially in systematic reviews and other evidence syntheses.

For example, an AI tool may help identify publications, screen abstracts, extract study characteristics, or prioritize records for review.

These activities may influence which evidence enters the final analysis.

When AI contributes materially to a documented search or selection procedure, researchers should report its role with sufficient detail for readers to evaluate the method.

Relevant information may include the tool used, the screening procedure, human validation, and limitations.

By contrast, using AI to suggest preliminary search terms may be treated differently under the applicable policy.

The requirements for disclosing AI-assisted literature searching therefore depend on the function the system performed.

Does AI-Assisted Research Data Coding Require Disclosure?

AI-assisted coding may materially affect how research data are classified or interpreted.

For example, a language model may assign preliminary categories to open-ended survey responses or classify documents according to a coding framework.

Such procedures can influence the findings, making methodological transparency important.

Researchers should explain the role of AI, how coding decisions were evaluated, and whether human reviewers corrected or confirmed the outputs.

The extent of reporting depends on the analytical procedure and journal requirements.

Researchers should also ensure that the processing of confidential or identifiable data complies with applicable ethical and data-protection obligations.

Disclosure after the fact does not retrospectively authorize an impermissible data transfer.

Does AI-Assisted Statistical Analysis Require Disclosure?

AI assistance with statistical analysis generally warrants careful consideration because it may affect the reported findings.

Examples include generating analysis scripts, recommending statistical tests, preparing data transformations, or interpreting model outputs.

The ICMJE identifies AI-assisted analysis as an activity that should be described in the methods section.

Researchers should document consequential AI involvement and verify the analytical procedures independently.

For example, generated code may execute successfully while using an inappropriate statistical model.

A methodological disclosure should describe what AI contributed without implying that the software independently validated the analysis.

The reporting requirements for AI-assisted statistical analysis should be considered alongside the researcher's responsibility for the results.

Does AI-Assisted Qualitative Analysis Require Disclosure?

Qualitative analysis often involves interpretive judgments that can be influenced by AI-generated suggestions.

A system may propose codes, identify recurring expressions, or generate preliminary themes from interview transcripts.

These outputs may be useful, but they can also overlook context, exaggerate patterns, or misrepresent participant meanings.

Where AI materially contributes to coding or interpretation, the methods section should explain how its outputs were generated and evaluated, according to the journal's requirements.

Researchers should not characterize substantive analytical assistance as simple manuscript editing.

They must also protect participant confidentiality and ensure that the use of external AI services is authorized.

Does AI-Generated Research Code Require Disclosure?

AI-generated code may be used for data cleaning, statistical analysis, simulation, visualization, or other research procedures.

The need for disclosure depends on the code's role and the applicable reporting policy.

Code that materially affects the analysis or findings may require methodological documentation, including the software environment, verification procedures, and relevant AI assistance.

By contrast, an AI suggestion for a minor formatting script may have limited methodological significance, although a publisher could still impose broader disclosure requirements.

Researchers should verify that the generated code performs the intended task correctly.

Reproducibility depends on the actual procedures and code used, not merely on naming the AI tool that suggested them.

Do AI-Generated Images or Figures Require Disclosure?

AI-generated or AI-modified images are subject to publisher-specific policies that may be more restrictive than those governing text.

Some publishers distinguish illustrative artwork from images representing research observations or evidence.

For example, AI-generated imagery should not be presented as authentic experimental, clinical, or observational evidence when it does not represent actual data.

Publishers may restrict certain generated images, require clear labeling, or impose additional permissions and documentation requirements.

Researchers should examine the intended journal's image-integrity policy before generating or submitting visual material.

Disclosure alone does not make prohibited image manipulation acceptable.

The requirements concerning AI-generated research imagery should be evaluated separately from ordinary language-editing policies.

What If AI Was Used for Several Different Activities?

A single research project may involve multiple forms of AI assistance.

For example, ChatGPT might help generate analytical code and later assist with manuscript language.

These uses should not automatically be combined into one vague statement because they may have different implications for methodological transparency.

Under policies that distinguish writing assistance from research methods, the researcher may need to report the analytical role in methods and the writing assistance in acknowledgments or a designated declaration.

Where a journal requires one consolidated statement, the different activities should still be described accurately.

The relevant question is whether the disclosure allows editors and readers to understand the actual contribution of AI.

Does the Extent of AI Use Determine Whether Disclosure Is Required?

The extent of assistance can matter, but quantity alone is not a reliable criterion.

A researcher may use AI repeatedly for minor punctuation corrections without materially affecting the research.

Another researcher may use AI once to generate a statistical procedure that determines the study's central findings.

The second interaction may be more consequential despite involving fewer prompts.

Disclosure decisions should therefore consider the nature and significance of the assistance, together with the applicable policy.

A simple count of prompts or AI-generated words does not adequately capture the role of AI in scholarly work.

Should Researchers Disclose AI Use That Did Not Affect the Final Work?

Not necessarily, unless an applicable policy requires broader reporting.

Researchers may experiment with AI suggestions that they later reject entirely.

For example, a model might propose an analytical approach that the researcher evaluates and decides not to use.

Such an interaction may not need to appear in the final manuscript if it did not materially contribute to the research and the governing policy does not require disclosure of exploratory use.

However, retaining a proportionate record of consequential decisions may be useful for the research audit trail.

Researchers should distinguish documenting an interaction from formally disclosing it in a publication.

Does Disclosure Make AI-Assisted Research Automatically Acceptable?

No. Disclosure does not override restrictions on AI use.

A journal may prohibit certain generated images, restrict substantive drafting, or require particular safeguards for AI-assisted analysis.

Similarly, an ethics approval may prohibit uploading identifiable participant data to an external AI service.

Researchers must establish that the activity is permitted and satisfy the relevant disclosure requirements.

They also remain responsible for verifying the accuracy and integrity of AI-assisted work.

Transparency is necessary where required, but it cannot substitute for methodological validity or ethical compliance.

What Should Researchers Do When the Policy Is Unclear?

Researchers should review the intended journal's author guidelines, publisher-level AI policy, and submission declarations.

They should also consider institutional and research ethics requirements.

If uncertainty remains about a consequential activity, a concise inquiry to the editorial office may clarify whether disclosure is required and where it should appear.

Researchers should preserve an accurate record of the assistance and any clarification received.

The absence of a detailed AI policy should not be interpreted automatically as permission to use AI without restriction or disclosure.

Which AI Activities Should Be Disclosed in a Mixed-Methods Research Project?

The researcher should document the consequential analytical assistance and explain how the generated outputs were checked.

The example illustrates why a single rule based on whether AI was used cannot capture the different implications of these activities.

The journal's actual policy remains decisive, and the same activities might receive different reporting treatment under another publisher.

Misconceptions About Which AI Activities Require Disclosure

How Should You Decide Which AI Activities to Report?

Begin by identifying the function of each AI-assisted activity and its effect on the research.

Then determine whether the activity is permitted and which disclosure requirements apply.

Researchers should also consider whether the activity involved confidential data or required prior institutional or ethical approval.

The appropriate disclosure is one that satisfies the governing policy and gives readers enough information to understand the AI contribution without exaggerating or concealing its significance.

What to Check When Classifying AI Activities for Disclosure

Frequently Asked Questions About AI Activities Requiring Disclosure

Disclose AI According to Its Role in the Research, Not Simply Its Presence

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.