In January 2023, the journal Science said text generated by AI tools could not be used in its papers at all. Within a year it had moved to a disclosure model, like most of publishing. The rules have kept shifting since, which is why so many researchers aren't sure what's allowed.
The short answer is that most journals let you use AI assistance, provided you say so, you check its output, and a human takes responsibility for every word. The longer answer depends on what you used it for and where you're submitting. We compared the policies of Springer Nature, Elsevier, Science, Wiley, Taylor & Francis and IEEE, together with the ICMJE recommendations that most medical journals follow and COPE's position on authorship.
This is a summary of publisher policies as of September 2026. Publishers revise these pages often, and individual journals can be stricter than their publisher. The journal's own guide for authors always wins.
Four things every policy agrees on
The wording varies, but on these points we found no major publisher that disagrees.
- AI can't be an author. Authorship means taking responsibility for the work, approving the final version and answering questions about it. A tool can do none of those. COPE, ICMJE and every publisher we looked at say this explicitly.
- You are responsible for everything AI produced. If a generated sentence is wrong, or a citation doesn't exist, that's the author's error, not the tool's.
- Substantive use has to be disclosed. Drafting text, rewriting passages and generating content all need declaring. The exemptions are for basic language checks.
- AI isn't a source. You can't cite a chatbot as evidence for a factual claim. ICMJE says so directly: these tools reproduce other sources, often inaccurately, without saying which.
Where they differ: publisher by publisher
The differences are in the details, and the details matter when you're filling in a submission form: where the disclosure goes, what counts as light editing, and whether you can use generated images.
| Publisher or body | Where to disclose | Light editing | AI-generated images |
|---|---|---|---|
| Springer Nature (incl. Nature) | Methods section, or a suitable alternative if there is none | AI-assisted copy editing needn't be declared; generative editing must be | Generally not permitted, with narrow exceptions |
| Elsevier | A declaration statement at the end of the manuscript, above the references | Basic grammar and spelling checks exempt | Not permitted, unless part of the research method and described there |
| Science (AAAS) | Cover letter and the manuscript itself | Check the current editorial policy; drafting text must be declared | Not permitted without the editors' explicit permission |
| Wiley | At submission, including how the output was reviewed and verified | Check the journal | Restricted; check the journal |
| Taylor & Francis | Acknowledged in the article, naming the tool and its purpose | Grammar and spelling tools exempt | Restricted; check the journal |
| IEEE | Acknowledgments, naming the system and the sections it affected | Grammar and editing help outside the policy, but disclosure recommended | Covered by the same disclosure rule |
| ICMJE (most medical journals) | Cover letter and manuscript: writing help in acknowledgments; data, analysis or figures in methods | Journal-dependent | Disclose in methods if used to produce figures |
Three patterns stand out. First, placement isn't standardized: the same use goes in the methods for one publisher and the acknowledgments for another. Second, the line for light editing differs. Springer Nature's copy-editing exemption is broader than a spell-check exemption, while IEEE recommends disclosing even routine language help. Third, images are treated far more strictly than text. Text can be checked by its authors; a generated micrograph or blot can't be, so most publishers simply don't allow them.
Which uses need disclosing?
Policies are written in general terms. Here is how they apply to the ways researchers actually use these tools, based on the policies above. Where a journal says something different, follow the journal.
| What you used AI for | Disclose? | Notes |
|---|---|---|
| Spelling and grammar checks | Usually no | Exempt under most policies; IEEE still recommends a mention |
| Rewording or paraphrasing your own sentences | Usually yes | Beyond copy editing once sentences are restructured or rewritten |
| Drafting paragraphs or sections | Yes | Every publisher requires this to be declared |
| Translating your own draft | Usually yes | More than copy editing under most definitions; check the journal |
| Searching for or screening literature | Yes, if it shaped which studies you included | Belongs in the methods, especially in systematic reviews |
| Writing analysis code or running analyses | Yes | Methods section, so others can reproduce the work |
| Summarizing papers for your own reading | Usually no | But you must check anything you then cite against the original |
| Generating images or figures | Usually not allowed | Narrow exceptions only; ordinary charts from your data are fine |
A disclosure statement you can adapt
A good statement names the tool, says what it was used for and in which sections, and confirms that the authors checked the output. Keep it specific. "AI was used in preparing this manuscript" tells an editor nothing.
During the preparation of this work, the authors used [tool name and version] to [purpose, for example: improve the language and readability of the Introduction and Discussion]. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.
Put it where your publisher asks: methods, acknowledgments or a separate declaration. Many submission systems also ask a direct question about AI use. Answer that consistently with your statement.
Why undisclosed use gets found
Since 2023, published papers have been found containing leftover chatbot phrases such as "Certainly, here is a possible introduction for your topic" and "as of my last knowledge update". Others have been caught with references that don't exist. These cases led to corrections and retractions, and to the stricter wording now in many author guidelines.
Detection software is a weaker risk than people assume. As we explain in our analysis of AI detectors, those tools are unreliable. The real risks are invented citations, overstated claims and text that no author can explain. Disclosing honestly and checking every claim protects you from all three.
If you're a peer reviewer
The rules for reviewers are stricter than for authors. A manuscript under review is confidential, and uploading it to a public AI tool can breach that. Elsevier and Springer Nature ask reviewers not to upload manuscripts to generative AI tools. The US National Institutes of Health has banned reviewers from using these tools on grant applications. If you want help wording a review, use it only on text you've written yourself, and check the journal's reviewer guidelines.
Students, theses and dissertations
Journal policies cover submitted manuscripts. Coursework, theses and dissertations are governed by your university, and often by the individual course. One module may allow AI for editing while another forbids it entirely. Check the policy for each assignment, ask your supervisor if it's unclear, and keep your drafts and notes so that you can show how the work developed.
Using LitraMind within these rules
LitraMind is built on the same principle the policies share: the author decides. AI Edit shows each suggestion separately, and nothing changes until you accept it. Research & Cite finds published sources for a claim, so your references point to real papers you can check.
Some features, such as AI Write and the rewrite modes, generate or restructure text. Under most of the policies above, that counts as use you should disclose. If you used them, say so in your statement. It's a sentence, and it's the honest answer.
Checklist before you submit
- Read the target journal's guide for authors on AI, not just the publisher's general page
- List every AI tool you used and what for
- Check every AI-assisted sentence, number and citation against your sources
- Remove any AI-generated images unless the journal explicitly allows them
- Write a specific disclosure statement and put it where the journal asks
- Answer the submission system's AI question consistently with that statement
- Make sure every co-author knows and agrees
