03 · What You Need to Know
Understanding AI Language-Editing Disclosure in Scholarly Publishing
Why do journals ask researchers to disclose generative AI use?
AI-use declarations can help editors and readers understand how generative systems contributed to manuscript preparation. They also clarify which activities were performed by human researchers and which received automated assistance.
Transparency matters because generative AI may introduce factual errors, alter scientific qualifications, misrepresent citations, or generate substantive interpretations. These risks are not identical across activities. Correcting punctuation is different from generating a theoretical explanation.
Disclosure requirements also reflect broader questions about accountability, reproducibility, and the provenance of scholarly content. However, a declaration does not establish that the disclosed assistance was appropriate or that the resulting manuscript is accurate.
Authors remain responsible for the final work, whether the assistance is reportable or exempt under a particular policy.
Is routine AI grammar correction the same as substantive language editing?
No. Routine correction generally concerns spelling, punctuation, grammatical agreement, and similar surface-level errors. Substantive language editing may involve extensive sentence restructuring, paragraph reorganization, changes in emphasis, or reformulation of arguments.
These activities can overlap. A request to correct grammar may produce a rewritten paragraph, and a request to improve clarity may introduce new scientific claims.
The relevant distinction therefore concerns the actual contribution rather than the label attached to the prompt.
| AI-assisted activity |
Nature of contribution |
Disclosure consideration |
| Spelling and punctuation checks |
Surface-level correction |
May be exempt under policies covering basic language checks |
| Grammar correction |
Correction of existing expression |
Check whether the journal exempts routine proofreading |
| Extensive sentence rewriting |
Substantial reformulation of prose |
May fall within reportable manuscript-preparation assistance |
| Paragraph restructuring |
Changes to organization and presentation |
May require disclosure even when scientific claims remain unchanged |
| Generating interpretations |
Substantive scholarly content |
Requires particular attention to permission, verification, and disclosure |
| AI-assisted research analysis |
Methodological or analytical contribution |
May require reporting within the research methods as well as other declarations |
This classification is a practical guide rather than a universal publisher standard. Researchers should apply the definitions adopted by their intended publication.
The distinction is examined further in AI grammar correction versus scientific content generation.
What does ICMJE recommend about AI disclosure?
The International Committee of Medical Journal Editors addresses the use of AI-assisted technologies in scholarly publishing. Its recommendations emphasize transparency about relevant AI use and the responsibility of human authors to examine generated material for accuracy, completeness, and appropriate attribution.
ICMJE also distinguishes AI assistance in writing from AI use that forms part of the research methodology. The latter may require description within the Methods section.
Importantly, ICMJE recommendations are not identical to the binding submission rules of every journal. Researchers should examine how the target publication implements those recommendations.
A journal may adopt additional requirements concerning the location, wording, or extent of an AI-use declaration.
How does Elsevier approach AI-assisted language editing?
Elsevier's generative AI policies distinguish basic language checks from more substantive generative AI assistance in manuscript preparation.
Under its updated policy framework, basic spelling, grammar, and punctuation checks are distinguished from substantive changes to sentence structure, organization, or content. The latter may require declaration even when the author regards the activity as language improvement.
This distinction illustrates why a researcher should not assume that every activity called proofreading is automatically exempt.
Policy language and journal-specific instructions should be checked at the time of submission, particularly when the manuscript has undergone extensive AI-assisted rewriting.
What about Springer Nature?
Springer Nature provides guidance concerning generative AI use in manuscript preparation, including distinctions between language improvement and more substantive AI-generated content.
Its approach emphasizes author responsibility, verification, and the need to ensure that scientific reasoning and interpretation remain under human control.
Researchers should examine the applicable publisher and journal instructions to determine which activities require acknowledgment or declaration.
A policy allowing AI-assisted language improvement should not be interpreted as unrestricted permission for AI to generate scientific findings or interpretations.
Does COPE require a particular disclosure statement?
The Committee on Publication Ethics provides guidance on authorship and AI tools, emphasizing that AI systems cannot meet the responsibilities associated with human authorship.
COPE guidance contributes to broader publication-ethics expectations, but researchers should not treat it as a universal manuscript submission form or a substitute for journal-specific disclosure instructions.
Its relevance lies particularly in maintaining human accountability and transparent scholarly practice.
Researchers should distinguish guidance issued by publication-ethics organizations from requirements imposed by individual publishers and journals.
Does disclosure depend on how much AI changed?
The extent of editing may matter, but word count or the percentage of revised text is not necessarily the determining factor.
One AI-generated sentence could introduce a substantive interpretation, while numerous minor punctuation corrections might leave the scientific content unchanged.
Conversely, extensive restructuring may be reportable under a policy even when the author has carefully preserved the underlying science.
Researchers should consider the nature of the activity, the significance of the changes, and the wording of the applicable policy.
The more detailed question of whether disclosure depends on the extent of AI editing deserves separate consideration because quantity and substance are not interchangeable.
What should an AI language-editing declaration contain?
Where disclosure is required, the statement should accurately describe the role of the tool without implying that it performed activities that did not occur.
Depending on journal requirements, relevant information may include the tool's name, its purpose, the type of assistance provided, and the authors' responsibility for reviewing the output.
For example, a researcher who used AI to improve grammar and sentence clarity should not describe the tool as having conducted research analysis. Conversely, substantive AI-generated discussion text should not be concealed beneath the phrase "minor proofreading."
Illustrative Declaration
Language-Editing Assistance
"During manuscript preparation, the authors used ChatGPT (OpenAI) to assist with language clarity and sentence-level editing. The authors reviewed and revised the suggested changes and take responsibility for the final manuscript."
This is an illustrative statement, not a universally approved disclosure template. The actual wording should follow the target journal's instructions and accurately reflect the assistance received.
Where should the disclosure appear?
Journals may specify an acknowledgment, a dedicated generative AI declaration, a Methods statement, or a submission-system disclosure field.
AI used in research analysis may need to be described in the Methods section because it affects how the study was conducted. AI used solely for manuscript language editing may be reported elsewhere, depending on the policy.
Researchers should not assume that mentioning AI in an acknowledgment automatically satisfies every disclosure requirement.
The journal's current instructions determine the appropriate placement.
Does disclosure make AI-generated scientific content acceptable?
No. Disclosure and scientific validity address different questions.
A declaration tells readers that AI assistance occurred. It does not establish that the generated statements are accurate, appropriately sourced, or permitted by the publication.
If AI introduced a fabricated citation, an unsupported interpretation, or an incorrect methodological description, the error must be corrected.
The obligation to verify AI-rewritten scientific text remains regardless of whether the assistance is disclosed.
Can researchers omit disclosure if they rewrite the AI-generated text themselves?
Manual revision does not automatically eliminate a disclosure obligation. Some policies concern the use of generative AI during manuscript preparation rather than the percentage of generated wording retained.
If AI contributed substantively to the organization or interpretation of a section, rewriting its output may not change the nature of the assistance.
On the other hand, an exploratory AI suggestion that was rejected entirely may be treated differently under a policy focused on material incorporated into the manuscript.
Researchers should apply the actual policy rather than assume that manual paraphrasing erases the history of assistance.
What if the journal's policy is unclear?
When instructions are ambiguous, researchers should review the journal's author guidelines, publisher policy, and submission declarations together.
If uncertainty remains and the AI contribution is substantive, clarification from the editorial office may be appropriate.
Keeping a concise record of the tool used, its purpose, and the changes accepted can make accurate reporting easier.
Transparency should describe the actual process rather than rely on vague or exaggerated claims about what the AI system contributed.
Watch Out
Do not assume that calling an activity "language editing" makes it exempt from disclosure. If AI substantially rewrote, reorganized, or generated scholarly content, check whether the journal treats that assistance differently from routine grammar correction.
