Key Takeaways
- Rules differ by publisher: Elsevier and PLOS exempt basic grammar fixes, while Springer Nature asks authors to describe even language editing.
- Four details form the statement: Name the tool and version, the task, the sections it touched, and how you checked the output.
- Placement follows the use: Research use goes in the Methods and figure work usually in the legend, while writing help goes in the Acknowledgments or, at Elsevier, a separate declaration.
- Check the journal's own rules: In 2023, 12 of 15 highly ranked journals whose publishers had AI guidance set rules that conflicted with it.
- The last line is a promise: Elsevier's template ends with the authors taking full responsibility for the content, claims included.
Asking a chatbot to improve the readability of your Discussion section needs a disclosure at Wiley and none at PLOS.
Each publisher draws the line where it will, and few submissions report crossing it. Of 25,114 research submissions to 49 BMJ Group journals between April and November 2024, 5.7% disclosed AI use. A separate estimate puts LLM processing in at least 13.5% of English-language PubMed abstracts from 2024, a lower-bound estimate.
An AI disclosure statement is a short passage in a manuscript that names the AI tool the authors used, says what it did and which parts of the work it touched, and explains how the authors checked its output. Journals that require one ask for it at submission, and the named authors stay responsible for everything it covers.
Authors who disclose well write the statement once, from the four details the major policies ask for between them. Then they fit it to the target journal's rules.
A statement that names the tool and version, the date of use, the task, the sections, and the check meets the core of what PLOS, Wiley, and Elsevier ask for. The target journal's author instructions decide where it goes.
Publisher Requirements Behind AI Disclosure
ICMJE's recommendations, updated in January 2026, tell journals to require AI disclosure at submission. Authors describe that use in both the cover letter and the manuscript.
The reason is accountability. ICMJE bars chatbots from authorship because they cannot be responsible for the accuracy, integrity, and originality of the work, so no tool can meet the authorship criteria. Disclosure keeps that responsibility with the named authors.
Journals have not settled on one rule. A 2024 audit published in The BMJ found that, by October 2023, 87% of the 100 highest-ranked science journals had a generative AI policy. Of the 15 journals whose publishers also had guidance, 12 set rules that conflicted with it. Funders run separate rules for grant applications, covered in our post on ethical AI use in proposal drafting.
researchers
Which AI Uses Need to Be Disclosed
Publishers agree at the ends of the range. Basic spelling and grammar checks need no statement at Elsevier or PLOS, and drafting text needs one at every publisher in the table, though Wiley sets the line at substantial text. The split comes in the middle, at language editing, readability, and flow, where Springer Nature asks authors to describe even grammar and language editing.
The table compares Elsevier, PLOS, Wiley, and Springer Nature, as of October 2026, on five uses, from grammar checks to AI tools for literature screening.
Taylor & Francis goes further and does not allow AI to write a first draft of a manuscript or any section of one. When use falls between the lines, Wiley advises erring on the side of disclosure.
What to Include in an AI Disclosure Statement
Between them, the major policies ask for four details. PLOS asks for all four apart from the version, and Elsevier, Wiley, and Taylor & Francis ask for most.
- Tool and version: PLOS, Elsevier, and Taylor & Francis ask for the tool's name. Wiley and Taylor & Francis also ask for the version. Elsevier asks for it in figure captions and the Methods, and Wiley asks for the date of use.
- Task: Say what the tool did, in verbs such as "suggested a reorganization" or "generated an R script." PLOS, Wiley, and Elsevier all ask how or why the tool was used.
- Sections affected: PLOS asks which parts of the study or article the tool affected. Wiley recommends naming the sections in Acknowledgments disclosures.
- How you checked it: PLOS requires a description of how the authors evaluated the output, and Elsevier asks authors to document the extent of their oversight. The check fits the use, such as rereading edited passages for meaning, rerunning code on the raw data, or confirming each suggested reference exists and supports its sentence. Our post on fabricated citations in biomedical papers explains why that last check matters.
If your journal follows WAME's 2023 recommendations to editors, also give the prompts used to draft new text. Keep a prompt log either way, as Elsevier advises.
Published statements may not include these details. In one study of 24 medical-education journals, 51 of 2,046 empirical articles from the first half of 2025 included an AI disclosure statement. Those statements averaged 43 words and described AI use superficially, most often as editing or translation.
Placing the Statement in Your Manuscript
Read the target journal's author instructions before manuscript submission, because journal rules can differ from their publisher's. Research use usually goes in the Methods, and writing help and figures vary more. The table compares where ICMJE, Elsevier, PLOS, Wiley, and Springer Nature place writing help, research or analysis use, and figures.
AI Disclosure Statement Examples by Use Case
Each example carries the four details, and brackets mark what you fill in. The writing-help examples follow Elsevier's template.
Language Editing and Readability
Use this version when a tool improved the readability or flow of your text. It goes in the Acknowledgments at ICMJE journals and Wiley, and in the separate declaration at Elsevier.
During the preparation of this work, the authors used [tool name, version] in [month, year] to improve the readability and flow of the Introduction and Discussion. The authors reread each edited passage against the original to confirm the meaning was preserved and take full responsibility for the publication's content.
Drafting Help on a Section
Check the journal's policy first, because Taylor & Francis journals do not allow AI to write a first draft. Otherwise, follow the journal's placement rules.
The authors used [tool name, version] in [month, year] to produce a first draft of the [subsection name] subsection of the Methods from protocol notes. The authors rewrote the draft, checked every step against the lab protocol, and take full responsibility for the content.
Literature Search and Screening
Research use belongs in the Methods. Place the statement next to the search strategy it supported.
We used [tool name, version] in [month, year] to screen [number] titles and abstracts retrieved from [databases] against the inclusion criteria in [section]. Two authors independently reviewed every record the tool excluded, resolved disagreements by discussion, and take full responsibility for the screening.
Analysis Code
PLOS requires disclosure of AI-generated analysis code. The statement goes in the Methods.
Analysis scripts in [language] were drafted with [tool name, version] in [month, year] and are available at [repository]. The authors reviewed each function, reran the full analysis on the raw data, confirmed the outputs against [independent check], and take full responsibility for the analysis.
Illustrative Figures
A schematic figure takes a disclosure in its legend, and Elsevier and Springer Nature also want it in the main statement. Springer Nature accepts it only when it derives from author-developed content or verifiable source material. Elsevier bars AI from creating or altering primary research images such as western blots.
Figure [number]. Schematic of the study design. The initial image was generated with [tool name, version] from the authors' draft schematic and edited by the authors, who checked every element against the protocol and take full responsibility for the figure.
Check the Claims Your Disclosure Vouches For
Elsevier's template ends with the authors taking full responsibility for the content of the publication. That sentence covers every claim in the manuscript, including passages no tool touched. The last check before submission is whether each claim holds up against its evidence.
QED Science's Paper Review helps with that check. It breaks a manuscript into claims, surfaces the gaps a reviewer would catch, and recommends fixes that the authors make. Reports stay in your account, and with author-controlled review, you decide whether to share them. If you are unsure whether your journal wants the review listed, name the tool and how you used it.
Before you sign the responsibility line, pull the manuscript's main claims into one list. Upload the manuscript to Paper Review to see where the evidence holds up for each one, and sign knowing which claims your evidence backs.
FAQs
researchers
researchers

