1. Introduction
Most journals now require authors to disclose their use of generative AI, but relatively few specify what an adequate disclosure should contain. As a result, many statements appear to fall short in one of two ways: they are too vague to be checked, or so detailed that reviewers may begin to question the extent of the author's own contribution.
This article summarises the points on which major policies agree, offers template wording for common situations, and describes two patterns that frequently prompt objections from reviewers. Because policies in this area continue to evolve, the specific journal's guidance for authors should always be consulted.
2. The point of consensus: AI cannot be an author
Although policies differ in detail, there is broad agreement that an AI system cannot meet the criteria for authorship (COPE, 2023; ICMJE, 2023).
The reasoning is consistent across these policies. Authorship entails accountability: the ability to approve the final version, to vouch for the integrity of the work, and to respond to questions about it after publication. An AI system cannot fulfil these responsibilities. Listing a chatbot as a co-author, which occurred in a small number of manuscripts in early 2023, is now likely to result in rejection.
A corollary is arguably more important than the rule itself: authors remain responsible for all content an AI tool contributes (COPE, 2023). Disclosure documents a process; it does not transfer responsibility. If an AI tool introduces an error into a manuscript, that error remains the author's.
3. Where the disclosure belongs
Placement varies between journals, and errors here are a common minor difficulty at submission.
- Language editing and formatting assistance is generally acknowledged in the acknowledgements section.
- Use that affects the research itself, such as data processing, coding, analysis, or literature screening, is usually reported in the methods section with the same specificity as any other research tool.
- Many publishers also require a declaration in the submission system or in a dedicated statement section. Elsevier, Springer Nature, Taylor & Francis, and Wiley each have their own requirements, so the target journal's guide for authors should be checked.
A useful general principle is that if the tool affected the evidence, the use belongs in the methods; if it affected only the prose, it belongs in the acknowledgements.
4. Template wording
The following templates address common situations. They are intended to be adapted rather than copied, and should name the actual tool and version used, since a phrase such as "an AI tool" provides little information to readers.
4.1 Language editing only
During the preparation of this manuscript, the author used Claude (Anthropic, model version [X], accessed [month year]) to improve the clarity and readability of the text. The author reviewed and edited all output and takes full responsibility for the content of the publication.
This is the most common case and the one on which publishers have been most explicit. Several publishers provide similar template wording, and using a publisher's own template is entirely acceptable and may be preferable.
4.2 Code generation for analysis
Analysis scripts in R were drafted with the assistance of GitHub Copilot ([version], accessed [month year]) and subsequently reviewed, corrected, and tested by the author. All statistical procedures were verified against [named reference implementation or worked example], and the analysis code is available at [repository link].
The verification clause is what makes this statement credible. Stating that generated code was checked, and against what, constitutes the substance of the disclosure.
4.3 Support for literature screening
Title and abstract screening was conducted independently by two human reviewers. [Model, version] was used as an additional screening pass to flag potentially relevant records that both reviewers had excluded; all flagged records were then assessed by a human reviewer against the pre-registered inclusion criteria. The model was not used to make inclusion or exclusion decisions.
The final sentence is particularly important. In evidence synthesis, reviewers are likely to focus on whether any model decision entered the results without human adjudication.
4.4 Assistance with qualitative coding
An initial codebook was developed inductively by the author from [n] transcripts. [Model, version] was subsequently used to suggest candidate codes for the remaining transcripts; all suggestions were reviewed against the transcripts by the author, and [n] were retained, [n] modified, and [n] rejected. Final coding decisions and all interpretation are the author's. Inter-coder agreement was calculated between the two human coders only.
This use tends to attract the closest scrutiny, which is understandable. Reporting how many suggestions were retained, modified, and rejected turns a general claim of oversight into a described procedure.
4.5 No AI use
The author declares that no generative AI tools were used in the preparation of this manuscript.
An explicit statement of this kind is increasingly required, rather than inferred from the absence of a disclosure.
5. Two common problems
5.1 Vagueness that cannot be verified
A statement such as "AI was used to assist with the preparation of this manuscript" illustrates the problem. It names no tool, version, task, or verification process, and could describe anything from spell-checking to substantial content generation. Readers learn only that some form of assistance occurred.
The remedy is straightforward: name the tool and version, the task, and the means of verification.
5.2 Over-disclosure that obscures the contribution
The opposite problem appears less common but may be more damaging. A statement listing every interaction across drafting, structuring, analysis, and interpretation can prompt a question the author did not intend to raise: what, precisely, was the author's own contribution?
A disclosure describes a process; it need not read as a confession. Reporting what the tool did, in terms similar to those used for any other research instrument, is usually sufficient.
6. Which tasks can be delegated
Beneath questions of wording lies a more substantive question about which tasks can appropriately be delegated. Although debate continues, the following distinction reflects much current guidance.
| Generally accepted with disclosure | Generally not acceptable, regardless of disclosure |
|---|---|
|
|
The organising principle is verification. A model may help to produce material that the author then checks; it should not produce material that is presented unchecked, because publication implies that a human author stands behind it.
7. A note on citations
Citations merit separate attention, because they appear to account for a considerable share of the corrections associated with AI use to date.
Language models can generate references that appear correct but do not exist. Such references may include plausible authors, titles, journals, and years, and occasionally a well-formed DOI that resolves to nothing or to an unrelated work. They are difficult to identify by inspection, because they closely resemble genuine references.
The most reliable safeguard is to verify every reference: resolve the DOI, confirm that the work exists, and confirm that it supports the claim attributed to it. This is time-consuming but necessary. A fabricated citation in a published article is likely to require a correction, and several may lead to retraction. Both outcomes remain part of the scholarly record and are largely preventable.
8. A checklist before submission
- Read the target journal's guidance on AI use, which may differ from that of previous journals.
- For each use, determine whether it affected the evidence (methods) or only the prose (acknowledgements).
- Name the tool, the version, the task, and the means of verification.
- Confirm that no AI system appears in the author list.
- Verify every reference independently of the tool that produced it.
- Keep a record of the tools used and how, so that questions can be answered accurately later.
9. Conclusion
Policies in this area are changing quickly, and specific requirements are likely to be revised. The underlying expectation, however, is stable and predates the technology: authors should describe their methods in sufficient detail for a competent reader to evaluate them, and should be able to stand behind every claim they publish.
References
- Committee on Publication Ethics (COPE). (2023, February 13). Authorship and AI tools [Position statement]. https://publicationethics.org/guidance/cope-position/authorship-and-ai-tools
- International Committee of Medical Journal Editors (ICMJE). (2023). Recommendations for the conduct, reporting, editing, and publication of scholarly work in medical journals. https://www.icmje.org/recommendations/