Using AI in your research paper: what to disclose and what journals expect
Yes, you generally have to disclose substantive AI use in a research paper. COPE says AI tools cannot be authors, authors are fully responsible for everything in the manuscript, and use of AI for writing, images or data analysis must be disclosed, usually in the methods, naming the tool and how it was used. Check each journal’s own policy before you submit.
Publishers have moved quickly. Almost every major journal now has a policy on AI, and the core rules are similar even where the wording differs. This guide summarises what the main bodies say, shows what a good statement looks like, and lists what to avoid. It is general guidance. The journal you submit to decides.
What do journals and ethics bodies agree on?
Three rules appear again and again in the COPE position on authorship and AI tools, in the ICMJE recommendations and in publisher policies:
- AI cannot be an author. An AI tool cannot take responsibility for the work, declare conflicts of interest or manage copyright.
- Authors are responsible. You are accountable for all content, including text, images or analysis that a tool helped produce, and for any breach of publication ethics.
- Use must be disclosed. COPE says to describe, in the methods or a similar section, how the tool was used and which tool it was. The ICMJE recommendations also ask authors to say in the cover letter and the submitted work whether they used AI-assisted technologies.
What do individual publishers say?
Publishers differ in detail. The table summarises points as described on a university library page updated in April 2025. Policies change, so read the current policy of your journal.
| Publisher | AI as author | Disclosure | AI-generated images |
|---|---|---|---|
| Springer Nature | Not allowed | Describe LLM use in the methods; routine copy-editing need not be declared | Generally not permitted, with limited exceptions |
| Taylor & Francis | Not allowed | A statement naming the tool and how it was used | Not permitted to create or alter images, figures or data |
| Wiley | Not allowed | Describe use in the methods or acknowledgements | AI may not create, alter or manipulate original research data |
| Elsevier | Not allowed | A declaration at submission, published with the paper | Not permitted to create or alter images |
Note the common pattern: language editing with a tool is often treated differently from using AI to generate text, analyse data or create images. Read the exact wording for your journal.
What should an AI-use statement say?
A good statement is short and specific. It names the tool and version, says what it did, says what you did to check it, and confirms that you take responsibility.
Template / exampleDuring the preparation of this work the authors used [tool name and version] to [specific purpose, for example: suggest a first draft of the related-work section / improve the clarity of the English]. The authors reviewed and edited the output, verified all facts, numbers and references, and take full responsibility for the content of the manuscript.
- Name the tool and version, and the date you used it if the journal asks.
- State the exact purpose. “To improve readability” is different from “to generate the introduction”.
- Say what you checked. The check is the part editors care about.
- Put it where the journal asks: usually methods, acknowledgements or a dedicated declaration.
What should I avoid?
- Listing an AI tool as an author or co-author.
- Using AI to create or alter figures, images or data unless the journal explicitly allows it.
- Trusting AI references. Tools can produce references that look real and do not exist. Verify every citation in the original source.
- Pasting unpublished or confidential material (yours, a co-author’s, or from peer review) into a public tool.
- Hiding the use. Disclosure is the safe choice, and not disclosing can lead to a rejection or a retraction.
What about AI detectors?
Detectors are unreliable and publishers differ in how they use them. Do not rely on one to decide whether your text is acceptable, and be wary of any promise that text will pass or fail a detector. The better protection is to write and check the content yourself, keep a record of what you used a tool for, and disclose it.
Keep a short record of your own contribution
Keep notes as you work: which tool, which task, which output you used, and what you changed. If a journal or supervisor asks, you can answer in one minute. The record also makes the disclosure statement easy to write.
When is disclosure clearly needed, and when is it a grey area?
| Use | Usually disclose? | Notes |
|---|---|---|
| Generating text for the paper | Yes | Name the tool, the purpose and what you checked |
| Creating or editing figures and images | Yes, and often not allowed | Many journals forbid AI-generated images; read the policy |
| Analysing data or writing analysis code | Yes | Describe it in the methods like any software |
| Summarising papers for your own reading | Often not required | You must still read and verify the sources |
| Grammar and spelling help | Varies | Some journals do not require it; others do |
| Brainstorming ideas | Varies | Check the journal and your university |
When the policy is unclear, disclose, or write to the editor and ask. A short email before submission is better than a long correction afterwards.
Protect confidential material
Do not put unpublished data, a co-author’s draft, or a manuscript you are reviewing into a public AI tool. Peer review is confidential, and many publishers ask reviewers not to upload manuscripts to tools that may store them. Read the tool’s data terms before you use it for anything private.
Where a tool can help
EasySeer creates only a first draft. When you set up the manuscript, you can choose to include an AI-use statement, and it helps you prepare the wording. You stay the author: you verify every result and reference, rewrite in your own words and disclose AI assistance as your university and journal require.
Results in the draft come from code that ran on your data, and weak results are reported as they are. EasySeer does not promise acceptance, a similarity score, or any outcome with an AI detector.
See the manuscript stageRead: ethics and AI disclosureEasySeer AI use and ethics
Questions
Do I have to disclose AI use?
For substantive use, such as generating text, creating images or analysing data, journals generally expect disclosure. Rules for minor language editing vary, so read your journal’s policy.
Can AI be a co-author?
No. COPE and major publishers state that AI tools cannot be authors because they cannot take responsibility for the work.
Where does the statement go?
Usually the methods or acknowledgements, or a declaration section. Some journals also ask for it in the cover letter. Follow the journal’s instructions.
Can I use AI to write the first draft?
Policies allow it with disclosure and human responsibility at many journals, but your university or funder may have stricter rules. Ask before you start, and verify everything the tool produces.
Do I need to disclose AI used only for grammar?
It depends on the journal. Some do not require it for basic language editing, and others do. Read the policy and, if unsure, disclose.
Keep reading
How to draft a manuscript
Choose the IEEE mode and tone, choose what to include, press Write manuscript, confirm that you have…
Guide · 2 min readHow to set up your manuscript
Before EasySeer writes the manuscript you choose the IEEE mode (Journal or Conference) and the tone …
Guide · 2 min readEthics and AI disclosure
EasySeer creates only a first draft from your concept. You stay the author: verify every result and …
FeatureManuscript: a first draft built from your own results
Choose the layout, citation style and tone, then EasySeer writes the manuscript from your concept, c…
Sources
Checked on 2026-10-24. Policies change, so read the current version before you submit.
- COPE: Authorship and AI tools
- ICMJE recommendations
- University of Edinburgh: what publishers say about AI (updated 16 April 2025)
- Springer Nature editorial policies
- Elsevier: generative AI policies for journals
Published 2026-10-24. Last updated 2026-10-24.