Yes, you can use AI tools like ChatGPT or Claude to help write a research paper in 2026 but with three firm limits: AI cannot be listed as an author, you remain fully accountable for every word, and any substantive AI use must be disclosed. Almost every major publisher now has a written policy, and many journals actively screen submissions for undisclosed AI. Getting the disclosure right is now part of getting published.
This guide explains exactly what the leading journals allow, what you must disclose (and what you don’t), how to write a compliant AI disclosure statement, and how to protect your manuscript from a growing problem: AI detectors that wrongly flag human writing, especially from non-native English speakers.
The three rules every major journal now agrees on
Despite differences in the fine print, Elsevier, Springer Nature, Wiley, Taylor & Francis, SAGE, the BMJ, ICMJE, and COPE have all converged on the same core principles:
- AI cannot be an author. Authorship carries responsibility for the work’s integrity, and a tool cannot take responsibility or approve a final manuscript. Every major publisher prohibits listing AI as an author or co-author.
- Humans stay accountable. You are responsible for the accuracy of anything AI helped produce, including any fabricated citations or errors it introduced.
- Substantive AI use must be disclosed. If AI contributed to the content, that use has to be transparent — the only real debate is where and how.
Keep these three in mind and you are already ahead of most submissions.
What you must disclose vs. what you don’t
Not every use of a tool needs a disclosure statement. The most useful distinction, articulated clearly in SAGE’s AI policy, is between assistive and generative AI:
| Type of use | Examples | Disclosure needed? |
|---|---|---|
| Assistive AI — refining your own words | Basic grammar and spell-check, light readability polishing that doesn’t change meaning | Generally no (varies by journal) |
| Generative AI — creating new content | Drafting sections, generating text, summarising literature into your paper, producing images | Yes — disclose, and often cite |
The line matters. Springer Nature, for example, specifically exempts pure AI-assisted copyediting from disclosure, while requiring that any substantive use of large language models be documented in the Methods section. When you are unsure, the safe move is to disclose over-disclosure is never penalised, but a hidden AI contribution can be treated as misconduct.
Where and how to disclose, by publisher
Disclosure placement has become more specific in 2026. Here is where the major players want it:
| Publisher / body | AI as author | Where to disclose substantive use | Notable specifics |
|---|---|---|---|
| ICMJE (medical journals) | Not allowed | Cover letter and manuscript; writing help → acknowledgments, data/analysis → Methods | Reviewers must not upload manuscripts to non-confidential AI tools |
| Elsevier | Not allowed | A dedicated statement placed before the references | Prohibits AI-generated or AI-altered images (except as documented methodology) |
| Springer Nature | Not allowed | Methods section | Exempts basic AI-assisted copyediting; prohibits AI-generated images |
| APA | Not allowed | Methods section, and cite the tool | Applies to AI used in drafting |
| SAGE / Wiley / Taylor & Francis | Not allowed | Varies; SAGE requires generative AI to be cited and referenced | Assistive AI generally exempt |
Because requirements differ not only between publishers but between journals within the same publisher, always confirm the exact wording in your target journal’s author guidelines before you submit.
How to write an AI disclosure statement (with examples)
A good disclosure statement is short, specific, and honest. It should name the tool, describe what it did, and confirm human oversight. Here are three adaptable templates.
1. Language and readability help only (assistive):
During the preparation of this manuscript, the author(s) used [Tool name, version] to improve the language and readability of the text. The author(s) subsequently reviewed and edited all output and take full responsibility for the content of the published article.
2. Generative help (drafting, summarising, restructuring):
During the preparation of this manuscript, the author(s) used [Tool name, version] to [draft sections of the introduction / summarise background literature / restructure the discussion]. All AI-generated content was verified, revised, and approved by the author(s), who take full responsibility for the final manuscript.
3. AI used in the research itself (data or analysis): Describe this in the Methods section, not just a disclosure line For example, the specific model used, the task it performed, and how outputs were validated.
Two practical notes: keep a short internal log of what you used AI for, in case an editor asks during review, and remember that if you used only a human editor or basic spell-check, no AI disclosure is required at all. That distinction is exactly why many authors prefer professional human editing for the language pass.
The non-native English trap: AI detectors and false flags
Here is the risk that catches careful, honest authors off guard. As journals adopt AI-detection screening, a well-documented flaw has surfaced: these tools disproportionately flag human writing by non-native English speakers as AI-generated.
In a widely cited Stanford study (Liang et al., published in Patterns), seven AI detectors incorrectly flagged 61.3% of TOEFL essays written by non-native English speakers as AI-generated, while classifying native-speaker essays with near-perfect accuracy. In that same set, 97.8% of the non-native essays were flagged by at least one detector.
The reason is technical, not a judgment of quality. Many detectors measure “perplexity” how predictable the writing is. Second-language writers tend to use more standard, textbook-correct sentence structures and vocabulary, which reads as “low perplexity” and trips the detector. Ironically, the cleaner and more conventional your English, the higher your risk of a false flag. Using automated grammar tools can increase that risk further, because they smooth out exactly the irregularities detectors associate with human writing.
For the international researchers who make up so much of global science, the takeaway is not to write worse English. It is to keep evidence of your authorship drafts, notes, version history and to be cautious about how your language is polished. This is precisely where human academic editing protects you: a qualified editor improves clarity without generating content, so there is nothing to disclose and a real person can vouch for the work. An editing certificate from a professional service also gives you documentary proof of a human language pass, which many journals now accept as part of their requirements.
A safe, compliant AI workflow for authors
You can use AI responsibly without risking your submission. A defensible workflow looks like this:
- Do your own thinking and writing first. Ideas, analysis, and conclusions should be yours — that is what makes you accountable.
- Use AI as a limited assistant, not a ghostwriter. Language polishing and brainstorming are lower-risk than generating whole sections.
- Never trust AI with references. Language models fabricate citations; verify every source against the original.
- Keep the confidential stuff offline. Do not paste unpublished manuscripts, peer-review reports, or patient data into public AI tools, which may retain your input.
- Log what you did, then write the disclosure statement to match.
- Check your target journal’s specific policy before submission; see our guidance on reading journal author guidelines to find these requirements quickly.
Common mistakes to avoid
- Assuming silence is safe. Undisclosed substantive AI use is treated as an integrity breach if discovered, and publishers are actively screening for it.
- Listing an AI tool in the author byline or acknowledgements as a contributor.
- Trusting AI-generated citations without checking them against the source.
- Pasting a confidential manuscript into a public chatbot.
- Over-relying on AI for the writing itself and losing the human accountability that authorship requires.
Frequently asked questions
Can I use ChatGPT or Claude to write my research paper? You can use them to assist — improving language, brainstorming, or restructuring — but not to author the paper. AI cannot be listed as an author, you remain fully responsible for the content, and any substantive use must be disclosed to the journal.
Do I have to disclose AI use if I only fixed grammar? Usually no. Basic grammar and spell-checking (assistive AI that refines your own words) is generally exempt, and many publishers, including Springer Nature, explicitly exclude AI-assisted copyediting. Generative use creating new text or content must be disclosed. Always check your specific journal.
Where do I put the AI disclosure statement? It depends on the publisher. Elsevier wants a dedicated statement before the references; ICMJE splits it between the acknowledgments (for writing help) and Methods (for data or analysis); Springer Nature and APA place it in the Methods section. Confirm the location in your target journal’s guidelines.
Will journals reject my paper for using AI? Not for disclosed, responsible use of assistive tools. Rejection risk comes from undisclosed use, AI-fabricated content or citations, or attributing authorship to AI not from transparent, well-documented assistance.
Why do AI detectors flag my writing when I wrote it myself? AI detectors are unreliable and biased against non-native English writers, falsely flagging a majority of such essays in independent studies. Keep drafts and version history as proof of authorship, and consider human editing over automated tools to reduce false-positive risk.
Conclusion
AI is now a normal part of the research writing toolkit but the rules are clear. Keep AI out of the author list, stay accountable for every claim, disclose substantive use in the place your journal specifies, and protect yourself against unreliable AI detectors. Do that, and you get the productivity benefits without putting your publication at risk.
If you want your language polished by a real human editor with nothing to disclose and a certificate to prove it. ManuscriptLab’s academic editing team can help your research read clearly while keeping you firmly in control of authorship.
