The short answer
Disclose any AI use that shaped the content or the research, name the tool and say how you used it and how you checked the output. You usually do not need to disclose spelling and grammar checks. AI can never be listed as an author, and you stay responsible for every word, number and citation.
This is not a trap. Publishers and the European Commission treat AI as a legitimate tool, the same way you would report statistical software. Disclosure is what lets readers trust your work.
What the EU guidelines ask for
The European Commission’s "Living guidelines on the responsible use of generative AI in research" (third version, May 2026) say researchers should name the generative AI tools they used substantially, and describe that use in the methods section (or equivalent) when it meaningfully shaped results.
They also explain what "substantial" means. Using AI as a basic editorial support tool is not substantial. Using it as a method or source, to interpret data, carry out a literature review, find research gaps, formulate aims or develop hypotheses could be. A good reference includes the tool name, version, date and how it was used. Where relevant, you can share prompts and outputs.
The guidelines also say plainly that AI systems are neither authors nor co-authors, and that the final responsibility for every citation stays with the researcher.
What the main publishers ask for
Policies are converging. All of the publishers below hold human authors fully responsible for the content, and most say explicitly that AI cannot be an author. They differ mainly in where the statement goes and what is exempt. Always check your target journal, since individual journals can add rules.
| Publisher | Where to disclose | Usually exempt |
|---|---|---|
| ICMJE (medical journals) | Writing help in acknowledgments; AI in data collection, analysis or figures in methods | Not specified |
| Elsevier | A separate declaration before the references; research use in methods | Basic grammar, spelling and punctuation checks, reference managers |
| Springer Nature | Disclose "evaluative" use such as suggesting methods or comparing results | Language polishing and structure are permitted (disclosure welcomed) |
| Wiley | Acknowledgments for writing help; methods for design, analysis or literature review; captions for images | Grammar, spelling, formatting, citation formatting |
| Taylor & Francis | A specific AI usage statement with tool name, version, purpose | Not specified |
| SAGE | Methods or acknowledgments for generative use | Assistive use: language, grammar, structure |
| PLOS | Methods, acknowledgments or figure legends, including how outputs were checked | Language improvements such as spelling, grammar and clarity (translation disclosure recommended) |
| IEEE | Acknowledgments, naming the sections and level of AI use | Editing and grammar (disclosure still recommended) |
Summarised from each publisher’s policy page in October 2026. Policies change, so check the current version before you submit.
What a good disclosure contains
Across the policies, a complete statement answers five questions.
- Which tool? Give the name and, if you know it, the version or model and the date you used it.
- For what? Drafting, editing, translation, literature search, screening, coding, analysis, code, figures.
- Where? Which parts of the paper or which stage of the research.
- How was it checked? For example, every reference verified against the original, every AI-suggested code reviewed by two authors.
- Who is responsible? A sentence that the authors take full responsibility for the content.
Example disclosure statements
Adapt these to your journal’s format. Replace the bracketed parts with your own details.
- Language editing (where disclosure is expected): "During the preparation of this manuscript, the authors used [tool] to improve the language and readability of the text. The authors reviewed and edited the output and take full responsibility for the content."
- Literature search: "We used [tool, version] on [date] to identify additional candidate studies. All suggested records were screened by two authors against the inclusion criteria, and only records verified in [database] were included."
- Qualitative analysis: "Interviews were transcribed with [tool], processed in the EU. Transcripts were checked against the audio by [author]. [Tool] suggested initial codes, which [authors] accepted, revised or rejected; themes were developed by the authors from the confirmed codes."
- Statistics or code: "[Tool] was used to generate analysis code in [language]. All code was reviewed, run and verified by [author], and results were checked against [method]."
- Grant application: "We used [tool] to help structure and edit parts of the project description. All ideas, aims and methods are our own, and all content was reviewed by the applicants."
Images, peer review and other special cases
Most publishers are stricter about images than text. Elsevier, Wiley and Taylor & Francis do not allow AI to create or alter primary research images such as microscopy or blots, and SAGE does not allow AI-generated images to be presented as original research images. Explanatory diagrams are often allowed if the tool is named in the caption.
Peer review is the other red line. The EU guidelines advise against substantial AI use in peer review and in evaluating funding proposals, and Elsevier, Taylor & Francis and SAGE tell reviewers not to upload manuscripts to AI tools. The reason is confidentiality: the manuscript is someone else’s unpublished work.
Will disclosing AI use count against me?
It is a fair worry. A 2024 study by Li and colleagues found that readers rated texts lower when AI assistance was disclosed, and a 2025 PNAS study by Reif, Larrick and Soll found that people who use AI at work are rated as less competent. The EU guidelines cite this kind of research and ask research organisations to avoid penalising honest disclosure.
The alternative is worse. Most publishers treat undisclosed AI use as a breach of policy, and ICMJE says nondisclosure may be treated as misconduct in some circumstances. A short, factual statement that shows you checked the output is the professional choice.
Keeping a record in Kahubi
Disclosure is easier when you have a record. In Kahubi every change to a draft, yours or the AI’s, is saved in the version history, screening decisions in a systematic review are logged with the AI suggestion and your final decision, and qualitative codes record what was suggested and what you confirmed. When you write the disclosure statement, the details are already there.
Sources
- European Commission (2026). Living guidelines on the responsible use of generative AI in research, third version, May 2026.
- COPE. Authorship and AI tools (position statement, 2023).
- ICMJE. Recommendations: artificial intelligence use by authors.
- Elsevier. Generative AI policies for journals.
- Springer Nature. Artificial intelligence (AI) editorial policy.
- Wiley. Using AI tools in your research.
- Taylor & Francis. AI policy.
- SAGE. Artificial intelligence policy.
- PLOS. Ethical publishing practice: artificial intelligence tools.
- IEEE. Author guidelines for AI-generated text.
- Li Z, Liang C, Peng J, Yin M (2024). How does the disclosure of AI assistance affect the perceptions of writing? EMNLP 2024.
- Reif JA, Larrick RP, Soll JB (2025). Evidence of a social evaluation penalty for using AI. PNAS 122(19).
- Luo X, et al. (2025). Reporting guideline for the use of generative AI tools in medical research: the GAMER statement. BMJ Evidence-Based Medicine.
- McAdoo T (2023). How to cite ChatGPT. APA Style blog.
Last updated 2026-10-09.