Is AI Allowed in Journal Submissions? Journal and Publisher AI Policies and How Much AI Is Acceptable in a Research Paper

Key Takeaways:

• AI assistance is allowed by almost every major publisher, but AI authorship is banned everywhere, and generative use must be disclosed.
• No publisher sets a permitted percentage: the test is whether the intellectual contribution is yours and whether you can be accountable for every claim.
• AI detector scores are unreliable, but journal editors and peer reviewers often can sense flat, boilerplate AI writing and become suspicious of your entire research.
• Disclose the tool, version, purpose, and extent of your oversight in the exact location your target journal specifies.

Glossary of Key Terms

Read this table first: the rest of the article uses these terms exactly as defined here.

Term	What it means in a publishing context
Generative AI	Technology that produces new text, images, audio, code, or synthetic data. Examples include ChatGPT, Claude, Gemini, Copilot, and DALL-E.
Large language model (LLM)	A statistical text model trained on large corpora. It predicts likely wording rather than retrieving verified facts.
Assistive AI	Tools that refine text you already wrote: spelling, grammar, punctuation, and readability suggestions. Sometimes exempt from disclosure.
Generative use	Any use where the tool produces substantive content or restructures your argument. Disclosure is required.
AI-assisted copy editing	Springer Nature’s term for readability, grammar, spelling, punctuation, and tone fixes to human-written text. Exempt from disclosure.
AI disclosure statement	A short declaration naming the tool, version, purpose, extent of use, and human oversight. Placement varies by publisher.
Hallucination	Plausible AI output that is factually wrong, including invented references, fabricated quotations, and non-existent DOIs.
AI detector	Software that estimates the probability that text was machine-generated. It measures style, not authorship.
False positive	Human-written text incorrectly flagged as AI-generated.
Perplexity	A measure of how predictable wording is. Low perplexity, common in careful academic prose, raises detector scores.
Human in the loop	The principle that a person reviews, verifies, and takes responsibility for every AI output.
Private AI tool	A tool that does not retain, reuse, share, or train on your input. Elsevier requires reviewers and editors to use only these.
COPE	Committee on Publication Ethics, whose position statement rules out AI authorship.
ICMJE	International Committee of Medical Journal Editors, which sets recommendations followed by most medical journals.
WAME	World Association of Medical Editors, which issued early recommendations on chatbots and scholarly manuscripts.

Is AI allowed in journal submissions?

Yes. Every major scholarly publisher now permits some AI writing assistance, and none permit AI authorship. What separates an acceptable submission from a rejected one is disclosure, verification, and how much of the thinking the tool did for you.

Policies from Elsevier, Springer Nature, Wiley, Taylor & Francis, Sage, BMJ, the JAMA Network, IEEE, and the Royal Society of Chemistry converge on 4 shared rules, then diverge on the details.

• Only humans can be authors, because only humans can be held accountable for a published work.
• Authors are fully responsible for accuracy, originality, and every reference, whatever tool produced the draft.
• Generative use must be declared: using AI for drafting paragraphs or sections, paraphrasing text, summarizing text, etc. Basic spelling, grammar, and punctuation checks are often exempt from declaration.
• AI must never fabricate, alter, or manipulate research data, results, or primary research images.

Beyond those 4 rules, requirements differ on where the statement goes, what counts as substantive, and how images are handled. The table below maps the current landscape.

What do the major publishers say about AI?

They all ban AI authorship and all require disclosure of generative use; they differ mainly on where the statement goes and how strictly images are controlled. Check the journal, not just the publisher.

Publisher	AI as author	Disclosure threshold	Key restriction
Elsevier	Not permitted	Declare when AI makes substantive changes to sentence structure or organization; basic grammar checks are exempt	No general-purpose AI images for graphical abstracts; no AI alteration of primary research images
Springer Nature	Not permitted	Declare generative use in the Introduction, Preface, or Acknowledgements; AI-assisted copy editing is exempt	Generative AI images and video are generally not permitted, with narrow labeled exceptions
Wiley	Not permitted	Declare in Methods or Acknowledgments when AI generates substantial text or restructures arguments	No AI creation or alteration of research data, results, photographs, micrographs, or diagnostic images
Taylor & Francis	Not permitted	Declare tool name with version, how it was used, and why, in a specific AI usage statement	No AI in the creation or manipulation of research results, clinical samples, or diagnostic images
Sage	Not permitted	Assistive use (spelling and grammar) needs no disclosure; generative use must be declared in Methods or Acknowledgements	Prohibits AI-generated participants, themes, data, images presented as novel, and fabricated references
BMJ	Not permitted	Declare in the contributor section, with a fuller description in Methods if AI was used in the research	May run screening tools before and after publication; inadequate declaration can trigger rejection
JAMA Network	Not permitted	Report in the Acknowledgment section, naming tool, version, manufacturer, and the sections affected	Prompts, their sequence, and ethics review details are required when AI forms part of the study
IEEE	Not permitted	Disclose AI-generated content in the Acknowledgments, identifying the system and the sections affected	Editing and grammar enhancement fall outside the core requirement, though disclosure is still recommended
Royal Society of Chemistry	Not permitted	Follows published principles of transparency, human oversight, and accountability; check journal guidance	Emphasizes governance and human oversight across editorial and publishing workflows

 

These documents are revised frequently: Elsevier updated its journal policy in June 2026, and BMJ commits to reviewing its policy at least every 6 months. Read the target journal’s instructions for authors on the day you submit.

How much AI writing is acceptable in a research paper

There is no permitted percentage. No major publisher sets a word count, a share of text, or a detector score as the limit. The test is qualitative: whether the intellectual contribution is yours, and whether you can defend every sentence.

A useful way to think about it: AI can help you say what you already know. It cannot create research for you. Elsevier states directly that AI tools must never substitute for human critical thinking, expertise, and evaluation.

The 3 tiers of AI use

Tier	Typical tasks	Disclosure status
1. Assistive	Spelling, grammar, punctuation, reference formatting, consistency checks	Generally not required
2. Generative or substantive	Drafting or rewriting sections, restructuring arguments, summarizing literature, translation, code generation, support for data visualization	Required, in the location your journal specifies
3. Prohibited	Fabricating data or references, generating results, altering primary research images, simulating participants	Not acceptable at any level of disclosure

 

A practical working limit

• Keep AI out of the Results as much as possible: these must come from your data and your judgment. If necessary, use AI only for language editing here.
• Draft the Introduction and Discussion yourself, then use AI to tighten what you wrote. Always supply PDFs or URLs of articles that you want to cite in these sections, and never ask AI to combine literature search and writing a review of the literature.
• Verify every reference in the original source; never trust an AI-supplied citation. Keep AI-assisted literature search separate from AI-assisted writing.
• Rewrite the key implications or arguments of your paper in your own voice so the manuscript reflects your analysis, not the model’s.
• Retain prompts, outputs, and dated drafts: several publishers may ask for documentation. Many AI tools provide permanent, shareable links of their output.
• If AI generates anything that you do not understand or cannot defend, it should not be in the paper.

What is the 30% rule in AI?

The 30% rule is an informal rule of thumb, not a policy. It usually means keeping an AI detector score below 30%. No publisher, standards body, or ethics organization recognizes it, and it appears in no instructions for authors.

The phrase circulates in 2 unrelated contexts. In student and forum discussions it describes a detector threshold. In business writing it describes a 70/30 split between automated and human work. Neither has any standing in scholarly publishing.

Why the 30% rule fails as a compliance test

• Detector percentages estimate style, not authorship, and different tools return different scores on identical text.
• A manuscript scoring 5% can still breach policy if undisclosed AI wrote the Discussion.
• A manuscript scoring 45% can be fully compliant if the writing is human, heavily edited, and properly declared.
• Editors judge intellectual contribution and transparency, not a number on a dashboard.
• Chasing a threshold encourages paraphrasing tools, which introduce errors and can themselves trigger integrity checks.

Will journals detect AI?

Often, yes, though rarely through detection software alone. BMJ states it may use screening tools before and after publication, and most publishers run integrity checks at submission. In practice, editors and reviewers usually notice content problems first.

How undisclosed AI use actually surfaces

• Automated integrity screening at submission, alongside plagiarism and image checks.
• Reviewer suspicion: Sage instructs reviewers who suspect undisclosed generative AI to raise it with the editor.
• References that cannot be located, which copy editors and production teams catch routinely.
• Numbers in the text that contradict the tables, figures, or abstract.
• Sophisticated prose that doesn’t say anything meaningful or hold together logically (called “AI slop” on social media).
• Reader reports after publication, which can trigger a correction, an expression of concern, or a retraction.

Disclosure rates suggest most AI use still goes undeclared. A cross-sectional study of 25,114 submissions to 49 BMJ Group journals found that only 5.7% disclosed AI use.

Are automated AI detectors accurate?

AI detectors currently are reliably enough to stand alone as evidence. Vendor claims and independent findings diverge sharply, and several universities disabled Turnitin’s AI detector over false positive concerns. Treat any score as a prompt to look closer, not as proof.

Many detectors measure perplexity: how predictable the wording is. Academic writing is predictable by design, so precise scientific writing and non-native English can both score as machine-like. This is also why AI-generated text can also appear on a plagiarism scan; much of academic phrasing is standard (e.g., “All participants gave written informed consent”).

What to do if you are wrongly flagged

• Ask which tool produced the score, which version, and what threshold was applied.
• Offer provenance: dated drafts, tracked changes, notes, outlines, and version history.
• Point to your disclosure statement if you did use AI within policy.
• Provide raw data, analysis scripts, and lab records where relevant.
• Ask the editor to assess the manuscript on verifiable content, not on a percentage.

Will I get flagged for AI if I use Grammarly?

Using Grammarly, Paperpal, or MS Word’s spellcheck for spelling, grammar, and punctuation is not a policy breach at any major publisher, and it normally needs no disclosure. Its generative rewriting features are treated differently and should be declared.

Where the line sits

Generally exempt from disclosure:

• Spelling, grammar, and punctuation correction
• Formatting of citations and references that you have provided the tool
• Consistency checks for tense, hyphenation, and reference formatting

Declare these:

• Full-paragraph rewriting, regeneration, or “improve it” style features (including Copilot in MS Word)
• AI-generated first drafts, outlines, summaries, or abstracts
• Translation of substantial passages
• Any output that changes meaning, structure, emphasis, or argument

Red flags for AI writing

Editors and reviewers look for patterns, not single words. The strongest signals are verifiable defects: references that do not exist, numbers that do not match the tables, and confident claims with no source behind them.

Red flags in the text

• Generic openings and closings that could introduce almost any field
• Confident assertions with no citation, or citations that do not support the claim made
• Uniform sentence rhythm and paragraphs of near-identical length
• Overuse of stock transitional phrases and hedging phrases
• Fluency that outstrips the depth of the analysis
• Leftover chatbot phrasing, such as references to a knowledge cutoff or to being a language model
• Abrupt shifts in voice between sections written at different times
• Mixed spelling conventions within a single manuscript

Red flags in the references

• Citations that cannot be located in any database
• DOIs that resolve to a different article, or that do not resolve at all
• Real authors paired with titles they never wrote
• Journal names that do not exist, or that are subtly misspelled
• Reference lists dominated by famous papers with none of the niche literature
• Publication dates that precede the method or technology being cited
• Page ranges and volume numbers that do not match

Red flags in figures, data, and methods

• Values in the text that contradict the tables, figures, or abstract
• Statistical tests named but not reported, or reported without assumptions or confidence intervals
• Methods described too vaguely to reproduce
• Images with implausible anatomy, labels, scale bars, or repeating textures
• Data visualizations that cannot be traced back to a stated dataset
• Sample sizes that change between the abstract, methods, and results

How do I declare use of AI?

Name the tool and version, state why and how you used it, describe how you verified its output, and confirm that you take responsibility for the entire paper. Place the statement where your target journal requires: Methods, Acknowledgments, or a dedicated AI section.

Where each publisher wants the statement

Publisher	Placement of the AI disclosure
Elsevier	A separate AI declaration section included at submission; the Methods section if AI formed part of the research methodology.
Springer Nature	Introduction, Preface, or Acknowledgements; the Methods section or equivalent for journal articles
Wiley	Methods or Acknowledgments, with a cover letter note at some journals
Taylor & Francis	A specific AI usage disclosure statement within the article
Sage	Methods or Acknowledgements, depending on where the tool was used
BMJ	The contributor section, with a fuller description in Methods if used during the research
JAMA Network	The Acknowledgments section. JAMA also has extensive guidelines on when AI was part of the study methodology, including discussing bias specific to AI in the Discussion section.
IEEE	The Acknowledgments section, identifying the affected sections of the article

 

What every disclosure should contain

• Full name of the tool, its version or model, and the developer
• Dates on which the tool was used
• The specific sections or tasks affected
• The reason the tool was used
• The extent of human review, editing, and verification
• An explicit statement that the authors take full responsibility for the content
• Prompts, settings, and outputs as supplementary material where the journal asks for them

Sample AI disclosure statement templates

Template 1, language and readability:

In preparing this manuscript, the authors used [tool name, version, developer] to improve language and readability. The authors subsequently reviewed and revised the text as required, and accept full responsibility for the content of the published article.

Template 2, literature search and synthesis:

The authors used [tool name, version, developer] on [dates] to identify and summarize literature relevant to the Introduction. Every source was independently located, read, and verified by the authors, who accept full responsibility for the accuracy of all citations and claims.

Template 3, AI used within the research:

[Tool name, version, developer] was used to [task], as described in the Methods. Prompts, model settings, and outputs are provided in the supplementary material. The authors verified all outputs and accept full responsibility for the analysis and conclusions.

Adapt the wording to your journal’s preferred format. Unless the journal specifies otherwise, keep the statement in the manuscript rather than only in the cover letter.

Do I need to declare use of AI for literature search?

In most cases, yes. Wiley’s guidance covers AI for literature review and data collection explicitly, and Sage lists assistance with literature review and compilation of sources as a disclosable use. Elsevier treats research-stage AI use as a Methods item.

• Declare it whenever AI selected, screened, ranked, extracted from, or summarized the sources you cite.
• Report it in the Methods section for systematic reviews and for any search intended to be reproducible. Although some tools claim that they can serve as the “second reviewer” required in a systematic review design, not all journals accept using AI in place of a competent human reviewer.
• Give the tool, version, dates, queries, databases, and screening criteria so the search can be repeated. Some journals ask for prompts and shareable links to be included in supplementary materials, so always save these.
• Verify every record in the original database: AI tools invent plausible but non-existent citations.
• Ordinary database search engines that happen to rank results algorithmically are usually treated as standard tools. As long as you didn’t use the AI component (e.g., you used Scopus but not Scopus AI), you don’t need an AI declaration for these.

Systematic reviews carry the highest burden. If AI touched screening or data extraction, PRISMA-style reporting expects that step to be described in enough detail to reproduce, including how disagreements between the tool and a human reviewer were resolved. Be sure to stay abreast about development of the PRISMA-trAIce Checklist (not currently registered on EQUATOR Network at the time of writing this article).

 

Can ChatGPT be listed as a co-author?

No. Elsevier, Springer Nature, Wiley, Taylor & Francis, Sage, BMJ, the JAMA Network, and IEEE all prohibit it, as does COPE. Authorship requires accountability, consent to publication, and approval of a final version. Only humans can provide those.

• Do not list an AI tool in the author byline or the contributor list.
• Avoid citing an AI tool as the primary source for a factual claim; cite the original literature instead.
• Every named author must still meet the journal’s and ICMJE authorship criteria independently of any AI use.

Can I use AI to create figures and images?

AI can be used for figures in some cases but never for research data. Publishers separate explanatory diagrams from research images. For example, Elsevier permits AI help with flow charts and schematics but prohibits AI-generated or AI alteration of primary research images.

Image type	Typical position	Requirement if permitted
Explanatory diagrams: flow charts, decision trees, timelines, schematics	Often permitted	Disclose the tool and version in the caption and in the general AI statement
Data visualizations: plots, charts, graphs, heatmaps	Permitted only when derived from the real underlying dataset	Describe the reproducible analytical method in the Methods section
Primary research images: microscopy, histology, blots, scans, patient images	Not permitted	Use established image processing software only; no AI adjustment of brightness or contrast
Graphical abstracts	Accepted by some publishers (e.g., Elsevier) but must comply with publisher’s AI use guidelines	Use dedicated scientific illustration tools rather than general-purpose image generators
Journal cover art	Case by case	Obtain prior permission from the editor and publisher, and clear all third-party rights

 

Springer Nature is stricter than most: generative AI images and video are generally not permitted for publication, with a narrow set of exceptions that must be legally sourced and clearly labeled as AI-generated. Confirm the rule before you invest time in artwork.

Can peer reviewers and editors use AI?

Only in a limited supportive role, and never by uploading the manuscript. Elsevier, Springer Nature, Taylor & Francis, and Sage all instruct reviewers and editors not to put unpublished manuscripts into generative AI tools.

If you’re peer reviewing a manuscript:

• Do not upload the manuscript, its figures, or any part of it into a public AI tool.
• Do not ask an AI tool to write, summarize, or evaluate the review report.
• Polishing the language of a review you wrote yourself is generally acceptable, with disclosure.
• Declare AI use in the review report, naming the tool and the purpose; basic spelling checks are exempt.

Elsevier asks reviewers and editors to use only private tools that do not retain, share, or train on inputs. Sage states that reviewers who generate reports inappropriately with AI will not be invited to review again.

What happens if you do not disclose AI use?

Consequences scale with severity, from an editorial query to rejection, correction, or retraction. Sage states that editors may reject a submission at any stage if inappropriate undisclosed AI use comes to light.

Situation	Likely outcome	Usual stage
Undisclosed minor language editing	Request to add or amend a statement	Copy editing or production
Undisclosed generative drafting	Desk rejection or a demand for major revision	Editorial screening or peer review
Fabricated or unverifiable references	Rejection and possible research integrity investigation	Any stage
Fabricated or altered data or images	Retraction and referral to the authors’ institution	Peer review or post-publication
Reviewer uploads a manuscript to a public tool	Removal from the reviewer pool and a confidentiality breach report	Peer review

 

Sage and its editors investigate concerns jointly, following COPE guidance and the publisher’s corrections and retractions policy. A disclosed use that an editor considers inappropriate is usually recoverable; an undisclosed one is far harder to defend.

Pre-submission AI compliance checklist

• Read the target journal’s instructions for authors on the day you submit, not from memory.
• Confirm whether that journal restricts AI to language improvement only.
• List every AI tool used, with version, developer, and dates.
• Map each use to a tier: assistive, generative, or prohibited.
• Verify every reference individually in the original source.
• Check every number in the text against the tables, figures, and abstract.
• Confirm no AI touched primary research images or underlying data.
• Write the disclosure statement in the exact location the journal specifies.
• Keep prompts, outputs, and dated drafts in case the editor requests documentation.
• Confirm that all listed authors are human and meet the journal’s authorship criteria.
• Check the co-authors’ contributions too: you are accountable for the whole submission.

Frequently Asked Questions

Can I use ChatGPT to write my research paper abstract?

You can use it to tighten an abstract you drafted, but generating one from scratch is generative use that must be declared, and some journals allow language improvement only. Abstracts are also where fabricated numbers do the most damage, since they are indexed, quoted, and read far more often than the full text.

Do I need to disclose AI use if I only used it for grammar and spelling?

Usually no. Elsevier, Springer Nature, Sage, and IEEE all treat basic grammar, spelling, and punctuation checks as exempt from disclosure. The obligation begins the moment the tool restructures sentences, reorganizes paragraphs, or drafts outlines or entire sections.

Will my paper be rejected if the AI detector score is high?

A score alone rarely causes rejection at a reputable journal, because detector output is contested evidence. Editors act on verifiable problems: fabricated references, unsupported claims, mismatched numbers, and flat, boilerplate writing that doesn’t say anything meaningful. Keep dated drafts so you can demonstrate provenance if a score is queried.

Is using AI in a research paper considered plagiarism?

Not automatically. Undisclosed generative use is normally handled as a transparency and authorship breach rather than plagiarism. It becomes plagiarism when the tool reproduces another author’s wording or ideas without attribution, which LLMs are known to do because they do not track sources reliably.

How do I cite ChatGPT in APA or Vancouver style?

Cite it as software, never as an author or as evidence for a claim. APA style formats a chatbot as a software reference: developer, year, model name with version, and a bracketed descriptor. Sage advises citing the original sources instead wherever possible, and reserving tool citations for generated images or visualizations.

Can I use AI to translate my manuscript into English?

Yes at most publishers. Taylor & Francis lists translation among supported uses, and Sage lists translation of research materials as a disclosable use. Declare it, then have a professional editor check technical terminology, because machine translation often mishandles discipline-specific vocabulary.

Do preprint servers require AI disclosure?

Policies vary by server, so the safe practice is to include the same disclosure statement you would submit to a journal. Journals will see the preprint during screening, and any mismatch between the 2 versions invites questions about transparency that are easy to avoid.

Can I use AI to help respond to peer reviewer comments?

Treat the response letter like the manuscript: AI may polish your wording, but the scientific reasoning must be yours. Do not paste confidential reviewer comments or an unpublished manuscript into a public tool, and declare substantive AI help in the revised submission if the journal asks.