03 · What You Need to Know
How to Determine Which AI Contributions Need Disclosure
Why Does AI Disclosure Matter in Scientific Research?
Disclosure helps readers, editors, reviewers, and other research stakeholders understand how scholarly work was produced.
When AI contributes to the methods or evidence, disclosure may be necessary for evaluating analytical validity, reproducibility, and the relationship between automated outputs and human judgment.
When AI assists with writing, disclosure may clarify the nature of the assistance and support compliance with publication policies.
However, disclosure is not evidence that an AI-assisted procedure was appropriate or that its outputs were accurate.
A manuscript can transparently report the use of AI while still containing methodological errors. Conversely, a scientifically valid procedure may be reported inadequately if the description omits consequential AI involvement.
Researchers should therefore distinguish transparency from validation and treat both as separate responsibilities.
Does Every AI Interaction Require Disclosure?
There is no single disclosure rule that applies to every AI interaction across all scholarly publishing environments.
Requirements depend on the governing policies and the actual contribution.
For example, a journal may require disclosure of generative AI used to produce manuscript text while treating ordinary spelling correction differently.
Another journal may require authors to identify any generative AI assistance used in preparing a submission.
Researchers should not assume that an activity is exempt merely because it appears minor. They should examine the policy applicable to the specific manuscript and submission date.
At the same time, the absence of a separate declaration requirement does not eliminate the need to describe an AI-assisted procedure that materially affects the research methodology.
Distinguish AI Disclosure From Methodological Reporting
AI-use disclosure
Identifies the involvement of AI-assisted technologies according to applicable transparency and publication requirements.
Methodological reporting
Explains how the research was conducted, including consequential AI procedures, inputs, outputs, validation, and limitations needed for scientific evaluation.
These forms of reporting may overlap but are not interchangeable.
A declaration that a language model assisted with qualitative coding does not explain how the coding was performed.
Similarly, a detailed methods section describing an AI classifier may not satisfy a journal's requirement for a separate AI-use declaration.
Researchers should establish which reporting functions apply and provide the relevant information in the appropriate manuscript locations.
When Should AI-Assisted Writing Be Disclosed?
Generative AI may assist with outlining, drafting, rewriting, summarizing, or editing manuscript text.
Disclosure requirements vary according to the publisher and the nature of the assistance.
The International Committee of Medical Journal Editors advises authors to disclose the use of AI-assisted technologies and describe how they were used, with appropriate reporting in the relevant manuscript sections.
Other publishers distinguish substantive generative AI assistance from basic spelling or grammar tools under their own policies.
Researchers should examine the current instructions of the target journal rather than assume that one publisher's exception applies universally.
Regardless of disclosure requirements, human authors remain responsible for the accuracy and integrity of the final text.
Does AI-Assisted Grammar Correction Require Disclosure?
Routine spelling and grammar correction may be treated differently from substantive text generation under some publisher policies.
However, the boundary is not always clear.
A tool that corrects punctuation performs a different function from one that rewrites an argument, introduces references, or changes the interpretation of results.
Researchers should examine what the system actually changed rather than rely solely on the label grammar assistance.
If an AI tool materially modifies scientific claims, the contribution may require disclosure or methodological explanation even when the original intention was editorial improvement.
The applicable journal policy should determine whether minor language assistance requires a formal declaration.
When Must AI-Assisted Literature Work Be Reported?
AI may suggest search terms, identify candidate publications, summarize retrieved articles, or extract information for evidence synthesis.
These activities differ in their methodological significance.
Using AI to brainstorm search vocabulary may be a relatively peripheral activity. Using AI to screen records or extract data for a systematic review can directly affect the evidence included in the study.
Where AI materially influences literature identification, eligibility decisions, or evidence extraction, the relevant procedure should be described sufficiently for methodological scrutiny.
Researchers should identify how original sources were verified and how errors or uncertain cases were handled.
Disclosure does not excuse fabricated references or unsupported summaries.
Should AI-Generated Statistical Code Be Disclosed?
AI-generated code may form part of a research procedure even when the final calculations are executed through conventional statistical software.
Researchers should distinguish code generation from computational execution.
For example, a language model may draft Python code, while the researcher selects the statistical method, inspects the script, executes it, and independently verifies the results.
When this contribution materially affects the analysis, the methods should provide enough information to explain the computational procedure and its validation.
Whether a separate AI declaration is required depends on the journal's policy.
Authors should not imply that AI independently conducted or validated an analysis when its role was limited to code development.
When Should AI-Assisted Data Cleaning Be Reported?
Data preparation can materially influence research findings.
If AI generates or applies rules for excluding observations, recoding variables, handling missing data, or combining records, these operations may require methodological reporting.
For example, an AI-generated script that removes incomplete responses affects the analytical sample.
Researchers should explain the relevant criteria, implementation, and verification rather than simply state that AI was used for data cleaning.
The scientific significance of the transformation determines how much detail is necessary.
When AI only suggests code that is subsequently reviewed and executed, that distinction should remain clear.
When Must AI-Assisted Qualitative Analysis Be Reported?
AI-assisted qualitative coding can influence the development of categories, themes, and interpretations.
Researchers should describe consequential contributions in a manner consistent with the selected qualitative methodology.
Relevant information may include the model, coding instructions, source material provided, human analytical decisions, and procedures used to evaluate generated suggestions.
For example, AI-generated preliminary codes should not be represented as themes independently established through researcher interpretation.
Likewise, a claim that all transcripts were analyzed manually would be misleading if automated classification materially shaped the results.
Reporting should make the relationship between AI assistance and human analytical judgment understandable.
What if AI Generates Research Measurements?
When AI produces variables, categories, scores, or classifications used in subsequent analysis, its role becomes part of the measurement procedure.
Such contributions generally require methodological description sufficient to evaluate the resulting evidence.
For example, a study using AI to classify students' engagement levels should explain the classification framework, relevant model information, validation procedures, and limitations.
Reporting only that an AI tool was used would not establish whether the measurements are suitable for the research question.
The necessary detail depends on the construct, research design, and consequences of classification errors.
Does AI-Assisted Research Design Require Disclosure?
AI may influence research questions, sampling decisions, instrument development, or methodological choices.
Not every exploratory suggestion requires a separate account in the final manuscript.
However, when AI materially shapes the final design or generates procedures that readers need to evaluate, the contribution may warrant disclosure or methodological reporting.
For example, an instrument containing AI-generated items should be described through its actual development and validation process.
Researchers should not present a measure as previously validated merely because AI recommended it or generated plausible items.
The key consideration is whether omitting the AI contribution would create a misleading account of how the study was conducted.
Which AI Uses Typically Require Different Forms of Reporting?
| AI activity |
Principal reporting consideration |
Likely location |
| Minor spelling or grammar assistance |
Check the publisher's specific policy |
AI declaration if required |
| Substantive manuscript drafting |
Identify generative assistance and preserve author accountability |
AI declaration or designated disclosure section |
| AI-assisted literature screening |
Explain screening, validation, and eligibility decisions |
Methods and any required AI declaration |
| AI-generated statistical code |
Explain consequential implementation and verification |
Methods and applicable declaration |
| AI-assisted qualitative coding |
Describe the analytical role, human decisions, and validation |
Methods |
| AI-generated research variables |
Explain measurement procedures and validity evidence |
Methods and limitations |
| AI-generated illustrations |
Follow image-generation, integrity, and labeling policies |
Figure information or required declaration |
| AI-assisted participant interactions |
Explain the intervention and applicable authorization |
Methods and ethics information |
This table is an illustrative reporting framework, not a universal publisher policy. The actual requirements should be checked for the target journal and research context.
How Much Detail Should an AI Disclosure Include?
A useful disclosure identifies the AI system and the specific nature of its contribution.
Where relevant, researchers may report the provider, model identifier, version information, date of use, and the tasks performed.
However, a declaration need not reproduce the entire analytical workflow when that information belongs in the methods section or supporting documentation.
For consequential procedures, sufficient methodological detail may include prompts, configuration, inputs, output handling, validation, and human oversight.
Researchers should distinguish information that is necessary for evaluation from incidental interaction history.
Confidential prompts or protected participant information should not be published merely to demonstrate transparency.
Should Researchers Disclose AI Use to Participants?
Participant-facing disclosure is different from journal disclosure.
When AI forms part of an intervention, consent process, or participant interaction, researchers should determine what information participants must receive under the applicable ethics and consent requirements.
For example, participants may need to understand that they will interact with an AI system rather than a human researcher.
However, not every use of AI in manuscript preparation creates a new participant notification requirement.
The appropriate communication depends on the approved research procedures, participant expectations, and governing obligations.
Researchers should avoid assuming that publication disclosure automatically satisfies participant-facing transparency requirements.
Does AI Use Need to Be Disclosed to Funders or Institutions?
Funding agreements, institutional policies, and research contracts may impose obligations separate from journal requirements.
For example, a project may restrict the use of external processing services or require approval of particular computational methods.
Researchers should examine these obligations before implementing the relevant AI activity.
A journal's permission to use generative AI does not override institutional confidentiality commitments or contractual restrictions.
Disclosure should therefore be evaluated across the relevant governance relationships rather than treated exclusively as a publication issue.
Can AI Be Listed as a Research Author?
AI systems do not meet the accountability requirements associated with scholarly authorship under ICMJE recommendations.
They cannot take responsibility for the accuracy and integrity of the work, approve the final manuscript in the manner required of human authors, or respond to scholarly accountability obligations.
Researchers should describe relevant AI assistance without presenting the system as an accountable author.
Human authors remain responsible for verifying generated content and correcting errors.
The fact that AI contributed substantial text or code does not transfer authorship responsibility to the tool.
What if the Journal Has No Explicit AI Policy?
The absence of a specific AI policy does not eliminate ordinary standards of methodological transparency and research integrity.
Researchers should still describe procedures necessary for understanding and evaluating the study.
If AI generated primary analytical variables, omitting that fact could leave the methods incomplete even without a separate disclosure requirement.
For uncertain writing-related cases, authors may seek clarification from the editorial office.
They should also follow applicable institutional or funding requirements.
When reporting is discretionary, a concise and accurate description of material assistance may be preferable to leaving consequential contributions ambiguous.
What if the AI Policy Changes During Manuscript Preparation?
Publisher policies may change between the beginning of a study and manuscript submission.
Researchers should consult the policy applicable at the relevant submission or revision stage.
Where requirements change, the research team may need to revise the declaration or provide additional methodological information.
The existence of a newer policy does not automatically establish that an earlier omission was deliberate misconduct.
However, authors should respond appropriately when a current requirement applies to their submission.
What if AI Use Was Not Disclosed?
An omitted disclosure should be assessed according to the actual AI contribution, the applicable policy, and the stage of publication.
Minor misunderstandings may require clarification or an updated declaration.
Omissions involving consequential analytical procedures may require more substantial methodological correction.
Where the manuscript has already been submitted or published, the appropriate response may involve contacting the editor and explaining the omission.
Researchers should distinguish an incomplete disclosure from evidence that the underlying research is invalid.
Nevertheless, a missing disclosure should not be ignored when it creates a misleading research record or violates an applicable requirement.
What Makes an AI Disclosure Scientifically Useful?
A useful disclosure is specific enough to explain the contribution without exaggerating its importance.
For example, saying that generative AI assisted with language editing communicates a different role from saying that an AI model classified interview responses used in the primary analysis.
The latter requires methodological information that a general acknowledgment cannot supply.
Researchers should identify what AI did, what humans retained responsibility for, and how consequential outputs were evaluated.
Transparency should enable readers to understand the research rather than merely demonstrate that a disclosure statement was included.
