Every major scholarly publisher now has a written position on generative AI use in manuscript preparation, and the individual mechanics — where a disclosure goes, what wording is required, which tools are exempt — are covered in detail on CASRAI’s AI disclosure guidance for authors. This page is the wider view: how those policies compare across the industry, where they converge, where they genuinely differ, and how the landscape has moved since publishers first started writing these rules. It’s about what authors submit to journals — a separate, related question is how reviewers and editors may use AI when handling a manuscript, which is governed by its own set of confidentiality-driven rules (see the AI in peer review Dictionary entry, and CASRAI’s separate comparison of publisher reviewer-side AI policies).
The one point every major publisher agrees on
No scholarly publisher or editorial body reviewed for this guide allows a generative AI tool to be listed as an author or co-author. The reasoning is consistent, not just the conclusion:
- ICMJE: AI tools “should not be listed as authors” because authorship requires taking accountability for the work, which a tool cannot do.
- Elsevier: “Authors should not list AI Tools as an author or co-author, nor cite AI Tools as an author,” because “authorship implies responsibilities and tasks that can only be attributed to and performed by humans.”
- Science / AAAS: “AI-assisted technologies such as large language models (LLMs), chatbots, and image creators do not meet the Science journals’ criteria for authorship and therefore may not be listed as authors or coauthors” — and sources cited in a Science paper may not themselves be AI-authored either.
This isn’t an arbitrary rule each publisher independently invented — it’s a direct extension of how authorship itself is already defined. ICMJE’s own authorship criteria require a substantial contribution, drafting or critical revision, final approval of the version to be published, and agreement to be accountable for the work. A generative AI tool can produce text or analysis, but it cannot agree to anything or be held accountable afterward. Every publisher’s “no AI authors” line traces back to that same accountability gap, whether or not the policy spells it out explicitly. See AI as author and AI co-authorship rejection (ICMJE 2023) for the Dictionary-level treatment of this specific question.
How the landscape got here: a rough timeline
Individual publisher policies didn’t appear all at once, and the space is still visibly moving. The general sequence, as far as it can be verified from publishers’ and trade bodies’ own dated material:
- 2021 — STM (the international trade association for academic and professional publishers) published an early framing document, “AI Ethics in Scholarly Communication,” well before generative AI tools were in mainstream use by authors.
- 2023 — The year most major manuscript-facing policies actually appeared. ICMJE clarified its position on chatbots and generative AI tools in its recommendations (a position it has reinforced in subsequent updates). Wiley circulated a formal “Generative AI Policy” for authors dated September 2023. STM followed in December 2023 with a considerably more detailed white paper, “Generative AI in Scholarly Communications: Ethical and Practical Guidelines for the Use of Generative AI in the Publication Process,” covering authors, editorial teams, peer reviewers, and technology vendors as four separate stakeholder groups.
- 2024 — Publishers moved from general statements to more specific rules. Taylor & Francis issued expanded guidance for authors, editors, and reviewers (dated June 2024), and publisher policies more broadly began drawing sharper lines around AI-generated images and figures specifically, not just AI-generated text.
- 2025 — STM’s Task and Finish Group on AI Labelling Terminology released a draft report aimed at classifying AI use in manuscript preparation more consistently across publishers — an attempt to reduce the current situation where each publisher describes similar AI uses in different words.
- 2026 — Two threads are running in parallel. STM published a further discussion document in March 2026, “Toward Responsible Use of Research Content in Generative AI,” though its focus is different from the author-disclosure question — it’s about how research content itself (the published record) should be handled by AI systems that consume it, not about what an individual author must disclose. Separately, a cross-organizational effort — the International Science Council, COPE, STM, the World Conferences on Research Integrity Foundation, and the Global Young Academy — is working toward a joint Global Reporting Standard for AI Disclosure in Research, with a content/taxonomy-identification phase running July–September 2026. It is not yet a published standard.
The trend across this timeline is convergence on the same handful of principles (no AI authorship, disclose substantive use, exempt basic tools) arrived at independently by different organizations, followed by a second wave of effort trying to standardize the terminology and mechanics rather than the underlying principles, which were mostly settled by late 2023.
What’s typically permitted vs. what’s typically prohibited
Reading across ICMJE, COPE, Elsevier, PLOS, Springer Nature/Nature Portfolio, Wiley, Taylor & Francis, and Science/AAAS policies together, a common shape emerges — with a few genuine points of divergence worth knowing about specifically.
| Use | Typical position | Notes |
|---|---|---|
| Listing an AI tool as an author | Prohibited, universally | No exceptions found across any policy reviewed. |
| Fabricating or manipulating data, images, or results with AI | Prohibited, universally | Treated as a research-integrity violation, not an AI-specific policy question — the same rules that already prohibit fabrication apply regardless of the tool used. |
| Basic spelling/grammar/punctuation checking | Generally exempt from disclosure | Elsevier explicitly exempts “basic checks of grammar, spelling and punctuation.” Most other policies draw a similar line, even where not stated in those exact words. |
| Substantive drafting, editing, translation, or summarization assistance | Permitted with disclosure | The core case every policy is written around — must be named and described, typically in the Methods, Acknowledgements, or a standalone declaration. |
| AI-assisted coding, data analysis, or literature search | Permitted with disclosure | Taylor & Francis names coding assistance and “interactive online search with LLM-enhanced search engines” as explicitly supported uses, provided they’re disclosed. |
| AI-generated images or figures depicting real results | Prohibited at several major publishers | Nature Portfolio journals prohibit AI-generated images in the manuscript, with narrow exceptions. This is one of the sharper divergence points: several biomedical/STEM-heavy publishers ban AI image generation outright rather than allowing it with disclosure. |
| AI-assisted conceptual illustrations (flow diagrams, teaching illustrations) not depicting research data | Permitted at some publishers, with disclosure | Taylor & Francis draws this narrower line explicitly: AI may assist with “research data visualisations and conceptual illustrations of processes, flow diagrams,” but AI use is not permitted for “the creation or manipulation of outputs of research or clinical testing” — a genuine, publisher-specific distinction, not a universal rule. |
The image/figure question is the clearest example of real divergence rather than cosmetic wording differences: a policy that bans AI-generated images outright and one that permits AI-assisted conceptual diagrams while banning AI-manipulated data images are drawing the line in genuinely different places, not just describing the same rule differently. Authors working across multiple target journals should check this specific point per publisher rather than assuming a single industry-wide answer — see AI image generation and Detection tool (AI-generated) for related Dictionary entries.
Where disclosure requirements diverge: placement
Every policy above requires disclosure of substantive AI use somewhere in the submission — but exactly where varies, and getting the placement wrong is a common, avoidable submission error:
- Elsevier: a standalone “Declaration of Generative AI and AI-assisted technologies in the writing process,” placed at the end of the manuscript, immediately above the references, and published in the final work.
- PLOS: “a dedicated section of the Methods, or in the Acknowledgements section for article types lacking a Methods section.”
- Science / AAAS: noted in both the cover letter and the methods or acknowledgments section, with the full prompt, tool name, and version disclosed in the methods.
- NEJM: required in both the cover letter and the manuscript body.
- Taylor & Francis: for articles, the Methods or Acknowledgments section; for books specifically, authors must disclose intent to use generative AI to their editorial contact at the earliest possible stage (proposal or writing phase), and, if approved, include the statement in the book’s preface or introduction — a distinct workflow from journal-article submission.
There is no single universal location. The practical implication is the same one that applies to every other author-facing publisher requirement: check the specific target journal’s instructions for authors rather than assuming the placement convention from a different publisher applies. CASRAI’s AI disclosure guidance for authors has ready-to-adapt sample wording for both a minimum-acceptable and a more substantive disclosure statement.
The common structure underneath the differences
Despite the placement and scope differences above, nearly every policy reviewed for this guide follows the same four-part shape:
- Authorship exclusion — AI tools cannot be listed as an author or co-author, and cannot be cited as one.
- Disclosure requirement — substantive use must be named: which tool, which version, what it was used for.
- An exemption for basic tools — spelling/grammar checkers, reference managers, and conventional statistical software (R, SPSS, Stata used in the ordinary way) are not treated as “AI use” requiring disclosure.
- Retained human accountability — the author(s) remain fully responsible for the accuracy, originality, and integrity of anything an AI tool contributed to, including verifying that the tool didn’t fabricate a citation or introduce an error.
If a new journal’s policy doesn’t fit neatly into this four-part shape, it’s worth reading it more carefully rather than assuming it’s equivalent to one you already know — a growing minority of policies (the image-generation restrictions above are the clearest example) add a fifth, publisher-specific element on top of this common base.
A separate question: AI used by reviewers and editors
The policies above all govern what an author submits. A different, adjacent set of rules governs what a peer reviewer or editor may do with a manuscript they’ve been asked to evaluate — chiefly a confidentiality question, since an unpublished manuscript uploaded into a third-party AI tool may expose unpublished data, methods, or findings to that tool’s provider. Taylor & Francis states this plainly for its own review process: “editors and peer reviewers must not upload files, images, or information from unpublished manuscripts into Generative AI tools.” This is a distinct policy question from author-side disclosure — worth knowing exists, but out of scope for this landscape page. See the AI in peer review Dictionary entry for the operational definition, and CASRAI’s separate publisher-by-publisher comparison of reviewer-side AI policies for the detailed reviewer-facing comparison.
Where this is headed
Two things are worth tracking rather than treating the landscape above as settled:
- The cross-organizational Global Reporting Standard for AI Disclosure in Research initiative (International Science Council, COPE, STM, the World Conferences on Research Integrity Foundation, and the Global Young Academy) is working through a taxonomy-identification phase in the second half of 2026, aiming at a shared vocabulary and structure for AI disclosure that individual publishers could adopt rather than each writing bespoke wording. It is not a published standard yet.
- STM’s ongoing work — the 2025 AI-labelling-terminology draft report and the March 2026 discussion document on responsible use of research content in generative AI — signals that publisher-facing guidance in this space is still actively being revised, not a one-time 2023 policy rollout that’s now finished.
Given the pace of change, treat any specific publisher policy detail — including everything on this page — as current as of when it was checked, and re-verify against the actual target journal’s current instructions for authors before submission. Publisher AI policies have already been revised multiple times since 2023 and there is no reason to expect that to stop.
Practical next step for authors
This page is the landscape view; two other CASRAI pages cover the practical mechanics once you know which journal you’re submitting to: AI disclosure guidance for authors has ready-to-adapt sample statements and a publisher-by-publisher placement summary, and How to disclose AI assistance in a CRediT statement, role by role covers the more specific question of how AI-assisted work interacts with a CRediT contributor statement, where one is required.
Frequently asked
Can a generative AI tool ever be listed as a co-author?
No, not under any major publisher or editorial body’s current policy. The consistent reasoning is accountability: authorship requires the ability to take responsibility for the work, which a tool cannot do, regardless of how substantial its contribution was.
Do I need to disclose using Grammarly or a basic spell-checker?
Generally no. Elsevier’s policy, for example, explicitly exempts “basic checks of grammar, spelling and punctuation” from requiring a declaration, and this exemption is consistent across most publisher policies even where not stated in identical wording. More substantive AI-assisted editing — restructuring passages, translating text, generating new content — does need disclosure.
Is one publisher’s AI disclosure statement automatically acceptable to another publisher?
Not reliably. Placement (Methods vs. cover letter vs. a standalone declaration vs. Acknowledgements) and required wording differ by publisher, and a statement written for one journal’s format may be missing an element another journal requires. Check the specific target journal’s current instructions for authors for each submission.
Does using AI to generate a figure or image violate publisher policy?
It depends on the publisher and what the image depicts. Several major publishers, including Nature Portfolio journals, prohibit AI-generated images in the manuscript outright. Others, like Taylor & Francis, permit AI-assisted conceptual illustrations (flow diagrams, process diagrams) with disclosure, while still prohibiting AI-assisted creation or manipulation of images depicting actual research or clinical data. There is no single industry-wide answer for this specific question — check per publisher.
Is AI use in peer review governed by the same policy as AI use by authors?
No. Author-facing AI policy governs manuscript preparation; a separate set of confidentiality-driven rules governs what reviewers and editors may do with a manuscript under review (for example, whether they may upload it to a third-party AI tool at all). See the AI in peer review Dictionary entry.
What happens if I don’t disclose AI use and a journal finds out later?
Consequences vary by publisher and severity, but nondisclosure is treated seriously rather than as a minor omission — ICMJE’s guidance states that nondisclosure “may require corrective action and may be construed as misconduct,” and several publishers reserve the right to decline or delay a manuscript over inappropriate or undisclosed AI use discovered during review.
