03 · What You Need to Know
What Makes an AI Disclosure Accurate, Informative, and Appropriate?
Why Is a Generic AI Disclosure Often Insufficient?
A statement such as "AI was used in this research" confirms the presence of assistance but provides little information about its significance.
The same sentence could describe routine grammar correction or an AI system that classified every participant response and generated the study's principal findings.
Those activities have different implications for methodological transparency and scholarly accountability.
A useful disclosure should therefore identify what the AI system actually did.
However, the appropriate level of detail depends on the journal's requirements. Researchers should not confuse a concise declaration with a complete methodological account.
Where AI materially influenced the research procedures, additional information may need to appear in methods even if a separate AI-use statement is provided.
What Do Major Publishing Authorities Require?
The International Committee of Medical Journal Editors (ICMJE) recommends that authors disclose the use of AI-assisted technologies and describe how they were used.
Its guidance distinguishes AI writing assistance from methodological AI use and emphasizes human responsibility for the resulting manuscript.
The Committee on Publication Ethics (COPE) similarly emphasizes disclosure of AI tools used in manuscript preparation and research activities, including information about the tool and its role.
Springer Nature and Elsevier provide publisher-specific requirements governing generative AI assistance and reporting.
These authorities share an emphasis on transparency, but they do not impose identical disclosure formats or universally require the same technical details.
Researchers should therefore follow the current instructions of the intended journal rather than treating one example statement as suitable for every publisher.
What Information Should an AI Disclosure Normally Consider?
The following elements provide a practical framework for preparing a disclosure. They are not a universal checklist imposed identically by every publisher.
| Information |
What It Explains |
When It May Be Needed |
| AI tool or provider |
Identifies the technology used. |
Commonly requested in AI declarations. |
| Model or version |
Clarifies which system produced the assistance. |
When required or relevant to reproducibility. |
| Purpose of use |
Explains why the tool was used. |
Central to an informative disclosure. |
| Activities affected |
Identifies writing, coding, analysis, translation, or other tasks. |
Needed to distinguish the nature of assistance. |
| Extent of contribution |
Clarifies whether AI suggested, generated, revised, or processed material. |
Especially relevant for substantive assistance. |
| Human verification |
Explains how outputs were reviewed or validated. |
Important for consequential research uses. |
| Methodological details |
Explains how AI influenced research procedures. |
When AI affects data, methods, or findings. |
| Prompts and settings |
Supports examination or reconstruction of AI interactions. |
Where required or methodologically significant. |
Researchers should include the details required by their journal and those necessary to explain consequential AI assistance.
A statement should not become longer merely to appear transparent. Its information should serve a clear reporting purpose.
Should the Disclosure Name the AI Tool?
Generally, yes, when the applicable policy requires identification of the technology used.
For example, a disclosure might identify ChatGPT as a tool provided by OpenAI or another AI application used during manuscript preparation.
However, naming the application is only the beginning of an informative declaration.
Different functions within the same application may have different research implications.
ChatGPT might be used to correct grammar, generate code, summarize publications, or suggest qualitative codes.
A disclosure that names the tool without explaining the activity may therefore be insufficient.
The researcher should identify the actual assistance rather than assume the application's name communicates its purpose.
Should the Model Version Be Included?
The model version may be relevant, particularly when AI materially contributes to research methods or reproducible computational procedures.
Different model versions can produce different outputs from similar inputs, and providers may update systems over time.
However, not every journal requires model-version information for every type of AI assistance.
For example, a brief declaration concerning language refinement may be subject to different requirements from a study systematically evaluating an AI model's classification performance.
Researchers should record the model information available at the time of use and report it when required or methodologically appropriate.
They should not invent a version number when the interface did not identify one.
The question of whether AI disclosures should identify the tool and model version is particularly important for studies involving AI as part of the research methodology.
Why Should the Purpose of AI Use Be Explained?
The purpose establishes why the technology was introduced into the research workflow.
For example, using AI to improve readability differs from using it to generate preliminary statistical code.
Both activities may be permitted, but they involve different forms of assistance.
A disclosure should identify the intended function clearly enough that readers can distinguish language support from methodological involvement.
Researchers should also avoid relying exclusively on the original prompt when the system performed additional tasks.
If a request for grammar correction resulted in substantive rewriting, the disclosure should reflect the assistance actually incorporated.
Should the Disclosure Identify Which Parts of the Manuscript Were Affected?
That may be useful or required, particularly when AI contributed substantive manuscript content.
For example, a researcher might report that AI assisted with language refinement in the introduction and discussion or generated preliminary code for the statistical analysis.
The appropriate specificity depends on the journal's policy.
Not every declaration needs to identify individual sentences or paragraphs.
However, a broad statement that AI assisted with writing may be misleading if the system also generated methodological descriptions or interpreted findings.
Researchers should distinguish the sections affected from the research activities performed.
A methods section written with AI assistance is not necessarily evidence that AI helped conduct the methods. Conversely, AI may influence analysis without generating any manuscript wording.
How Should Researchers Describe the Extent of AI Assistance?
The extent of assistance should be described in terms of what the system contributed rather than an arbitrary percentage of AI-generated text.
For example, a researcher might state that AI suggested preliminary codes, generated an initial script, or assisted with revising manuscript language.
These descriptions communicate the function performed.
Statements such as "AI contributed approximately 20% of the manuscript" may be difficult to interpret and verify unless the percentage has a clearly defined basis.
A word count does not establish the intellectual or methodological significance of AI assistance.
A single generated statistical procedure may influence the entire analysis, while hundreds of grammatical corrections may leave the substantive research unchanged.
Researchers should therefore describe the contribution rather than rely on unsupported numerical estimates.
Should Human Review and Verification Be Mentioned?
Human verification is often important, particularly when AI generated substantive claims or contributed to research analysis.
A disclosure may explain that researchers checked references, tested generated code, reviewed coding suggestions, or evaluated interpretations against the original evidence.
However, these statements must accurately reflect what occurred.
A researcher who merely read an AI-generated paragraph should not claim that every factual statement was independently verified.
Similarly, executing generated code without examining its logic should not be described as comprehensive statistical validation.
Verification language should identify actual procedures rather than provide a generic assurance of human oversight.
The separate question of documenting how AI outputs were verified becomes particularly important when AI materially affects published findings.
Is Saying "The Authors Take Full Responsibility" Enough?
No. Such a statement may accurately express accountability, but it does not explain the assistance provided or establish that appropriate verification occurred.
Major publishing authorities require human authors to remain responsible for submitted work regardless of AI involvement.
That responsibility exists even when the manuscript contains no explicit sentence asserting it.
A useful disclosure should therefore communicate the actual AI activity.
For example, stating that ChatGPT generated preliminary analytical code and that the researchers tested it provides more information than a declaration consisting only of a general responsibility statement.
Accountability language may complement the disclosure, but it should not replace an accurate description of the workflow.
Should the Disclosure Include the Exact Prompts?
Not universally. Some journals require only a description of the AI tool and its role.
Prompts may become methodologically relevant when they determine how AI processes research data or produces outputs central to the findings.
For example, a study using a language model to classify interview excerpts may need to document the instructions that defined the coding task.
In such cases, prompts may be reported in methods, supplementary materials, or a research repository, subject to applicable confidentiality restrictions.
However, a manuscript-language disclosure does not necessarily require the complete history of every editing conversation.
The question of whether researchers should disclose AI prompts depends on the journal's requirements and the prompts' significance for understanding or reproducing the procedure.
Should AI-Generated Responses Be Included in the Disclosure?
Not necessarily. An AI-use declaration ordinarily summarizes the assistance rather than reproducing every generated response.
However, outputs may need to be preserved or reported when they constitute research data, analytical results, or materials necessary for methodological evaluation.
For example, a study examining the accuracy of AI-generated feedback may need to retain the actual responses analyzed.
Likewise, a researcher using AI to generate preliminary qualitative codes may need to preserve those suggestions and subsequent revisions.
The distinction is between disclosure of the assistance and documentation of the underlying research procedure.
Researchers should follow applicable data-retention and sharing requirements while protecting confidential information.
What Should Be Reported When AI Contributes to Research Methods?
Methodological AI use may require more detail than assistance limited to manuscript language.
For example, if AI classified research responses, the manuscript may need to describe the classification procedure, input materials, coding framework, and verification undertaken.
If AI generated statistical code, readers may need information about the analytical procedure, software environment, and how the results were validated.
Where AI-assisted translation affected research instruments or interview transcripts, the translation and review procedures may need to be explained.
These details are necessary because they concern how evidence was generated or interpreted.
A short declaration identifying the tool may satisfy one reporting requirement without providing a complete methodological account.
How Do Writing and Methodological Disclosures Differ?
AI writing disclosure
Explains how AI assisted with preparing or revising the manuscript, such as drafting, language refinement, or substantive rewriting.
Methodological AI disclosure
Explains how AI contributed to research procedures, data processing, analysis, or interpretation, including information necessary to evaluate the resulting findings.
The distinction is important because the same AI system may perform both functions during one project.
Researchers should describe them accurately rather than combining consequential analytical assistance with routine language editing.
The appropriate placement is discussed further in where AI use should be disclosed in a research manuscript.
Does AI-Assisted Grammar Correction Require the Same Information?
Not necessarily. Some publishers exempt defined routine AI-assisted copyediting from declaration.
Where disclosure is required, a concise description of language assistance may be sufficient.
For example, the researcher might identify the tool and explain that it was used to improve grammar and clarity in independently written text.
However, the statement should not describe substantive drafting as routine copyediting.
The actual assistance and publisher policy determine whether a declaration is necessary and how detailed it should be.
Researchers should not assume that a complex methodological AI disclosure is required for every minor language correction.
What If AI Was Used for Several Different Activities?
A single statement may need to distinguish multiple activities.
For example, ChatGPT may have assisted with literature exploration, generated preliminary analytical code, and refined manuscript language.
These functions should be identified separately enough to communicate their different contributions.
Where the journal requires a consolidated AI-use declaration, the researcher can summarize the activities and refer to fuller methods reporting where appropriate.
The declaration should not imply that all AI use was limited to manuscript preparation if the system also influenced the analysis.
Consistency between the AI statement, methods section, acknowledgments, and submission declarations is essential.
Should the Disclosure Identify When AI Was Used?
The date of AI use may be relevant when a model or service changes over time or when the AI interaction forms part of a reproducible research procedure.
For example, a study evaluating AI-generated classifications may need to identify when the system was accessed.
However, not every publisher requires exact dates for routine writing assistance.
Researchers may benefit from recording dates internally so they can reconstruct the workflow if questions arise later.
Where dates are reported, they should reflect actual records rather than estimates presented as precise facts.
The separate issue of recording the dates of AI use concerns research documentation as well as publication transparency.
Should the Disclosure Describe AI Limitations?
Limitations may be relevant when AI materially affects the research methodology or findings.
For example, an AI classifier may perform unevenly across categories, or an AI translation system may have difficulty preserving culturally specific expressions.
These limitations may need to be discussed in the methods or limitations section rather than incorporated entirely into a short declaration.
Researchers should distinguish general limitations of generative AI from limitations demonstrated or reasonably relevant to the specific procedure used.
A generic statement that AI may produce errors does not substitute for evaluating the actual analytical risks.
The purpose is to explain uncertainties that matter for interpreting the research.
Can an AI Disclosure Be Too Detailed?
Yes. Excessive detail may obscure the information readers need, particularly when the disclosure reproduces lengthy prompts or conversational histories unrelated to the final work.
For example, a researcher who used AI only for permitted language refinement may not need to publish every editing request.
Conversely, a researcher who used AI to classify thousands of observations may need substantial methodological documentation.
The appropriate level of detail depends on the significance of the assistance, reproducibility requirements, and publisher instructions.
Researchers should separate concise disclosure from supporting research records where appropriate.
Transparency is not necessarily improved by including information that is irrelevant, misleading, or impossible for readers to interpret.
What If the Researcher Does Not Know the Model Version?
Researchers should report information they can substantiate.
Some AI interfaces do not clearly identify the underlying model version, and providers may change model implementations without displaying detailed technical information.
If the journal requires version information, the researcher should consult available records or seek clarification about acceptable reporting.
They should not invent a model identifier or infer a precise version from the current interface.
Where appropriate, the disclosure may identify the application and explain that the exact underlying model version was not available.
Accurate acknowledgment of unavailable information is preferable to false precision.
Does a Disclosure Statement Prove That AI Use Was Ethical?
No. Disclosure and ethical compliance are separate matters.
A researcher may fully disclose the use of AI to analyze participant interviews while having uploaded confidential data to a service without the necessary authorization.
The declaration does not retrospectively resolve that problem.
Similarly, an AI-generated image may be prohibited by a publisher even if it is accurately labeled.
Researchers must establish that the assistance was permitted and complied with applicable ethical, institutional, contractual, and publisher requirements.
The disclosure should accurately describe the assistance, but it cannot substitute for authorization or methodological validity.
What If the Journal Provides a Required AI Disclosure Template?
Researchers should follow the journal's prescribed format.
Some publishers specify a heading, statement structure, or location for generative AI declarations.
Others require information through an online submission form.
Where a template is provided, the researcher should adapt it to the actual assistance rather than inserting generic language that does not accurately describe the work.
For example, a statement claiming that AI was used only for readability would be misleading if the system also generated analytical code.
Researchers should also avoid adding unsupported verification claims simply because they appear in an example declaration.
What If the Journal's Disclosure Requirements Are Ambiguous?
Researchers should consult the journal's current author instructions, publisher-level AI policy, and submission declarations.
If uncertainty remains, a concise inquiry to the editorial office may clarify which information is required.
The inquiry should describe the actual AI activities and ask whether additional details are needed.
For example, a researcher might explain that AI generated preliminary statistical code that was subsequently tested and ask whether the journal requires the model version or prompts.
Such a question is more useful than asking whether ChatGPT is generally permitted.
The researcher should retain an accurate record of the assistance and any clarification received.
