Do you need to disclose AI use in a research paper? Springer Nature guidance and examples
If you use AI to prepare a manuscript for a Springer Nature journal, describe what it did and confirm that the authors checked the output and take responsibility for the paper. The publisher's current manuscript-preparation guidance asks for a declaration even for low-risk uses such as language editing. Check the target journal's instructions and submission form before deciding what to include and where to put it.
On November 5, 2026, Nature Geoscience published an editorial on responsible AI use. It explains the publisher's risk-assessment approach and the journal's updated policy pages. The broader framework was announced on September 10, 2026; it was not first introduced by the November editorial.
Policy sources checked November 8, 2026. This guide covers the linked Springer Nature guidance. Other publishers and individual journals may have different requirements.
Does language polishing need an AI declaration?
Under the current manuscript-preparation guidance, grammar editing, translation, formatting and summarising author-written text are green, low-risk uses when they do not alter scholarly content. The declaration column still says to clearly describe the use and confirm author accountability.
An old recommendation that language polishing never needs disclosure can therefore lead you to omit information requested by the current guidance. Record the actual task. If a tool also suggested interpretations or wrote new paragraphs, a statement saying it was used only for grammar would be incomplete.
What do the green, amber and red categories mean?
| Category | Examples in the guidance | What authors should do |
|---|---|---|
| Green: low risk | Language editing, translation, formatting, summarising the authors' text without changing scholarly content | Check accuracy and meaning, describe the use, and confirm responsibility |
| Amber: requires care | Outlines, drafts based on author input, reference-management support, visual content based on verifiable material | Verify content and citations, identify AI contributions clearly, and describe human review |
| Red: not permitted | Replacing authorship or independent scholarly judgement; invented empirical data or figures; listing AI as an author | Do not submit this use as acceptable merely because it is disclosed |
The same tool can be used for different tasks. Classify what happened to the manuscript, not the tool's brand. A declaration does not make a prohibited use permissible.
For figures, the policy distinguishes visual content derived from verifiable inputs from opaque generated content. Inspectable source data, author-developed material or validated computational outputs matter. AI-assisted visual content needs appropriate disclosure in the manuscript and specific information in the caption or legend. Keep the source data, code and review record available.
Record the task, review the output, and describe the actual AI contribution and author responsibility.
Keep an AI-use record while you write
Use one row per task. This is a working record for the authors, not a prescribed Springer Nature form. Dates and file identifiers help reconstruct the work even when the journal does not request them.
| Field | What to record |
|---|---|
| Tool and model | Product name, model/version shown, provider; if no model is shown, record that rather than guessing |
| Date and location | Date, manuscript section, figure number or source file |
| Purpose and input | Language editing, outline, code assistance or another specific task; which author material was supplied |
| Contribution retained | What was accepted, rewritten or discarded |
| Human verification | Who checked facts, references, numbers, meaning, code or visual output, and how |
| Evidence retained | Source version, relevant prompts/output, data and code where appropriate |
For example: an AI tool rephrased the abstract, but the authors rejected a sentence that broadened the study population. Record the accepted changes and the correction. Before/after source files are more informative than a note saying “AI improved the paper.”
Three AI disclosure statement examples
These are editorial examples, not official publisher templates. Replace every bracketed field with accurate details and remove claims that do not match your work. For an English-language submission, wording such as the following can be a starting point.
Language editing
During manuscript preparation, the authors used [tool, provider and model/version if available] to edit the language of [sections]. The authors compared the suggestions with the original text and checked that the scientific meaning, numerical values and citations were preserved. The authors reviewed the final manuscript and take full responsibility for its content.
Use this only when it describes the actual scope. Translation or summarisation should be named if it also occurred.
Drafting from author-supplied material
The authors used [tool and model/version if available] to propose an outline and draft text for [sections] from [author-prepared notes/material]. The authors checked each retained claim against [sources/data], verified the references, and revised the text. The research interpretation and conclusions were determined by the authors, who take full responsibility for the manuscript.
Do not claim to have verified references until someone has checked the original records and whether they support the surrounding claims. The citation-checking guide separates bibliography consistency from checking the evidence itself.
A figure produced from verifiable data
The authors used [tool and model/version if available] to assist with code for Figure [number] using [dataset/source identifier]. The authors reviewed and executed the code, compared the plotted values with the source data, and checked labels and units. The data and code are available at [location]. The authors take responsibility for the figure.
This example covers code-assisted plotting, not an invented experimental image. Only state that data and code are available if they actually are. Add a figure-caption disclosure identifying the tool, purpose and extent of its contribution, adapted to the journal's requirements.
Where should the declaration go?
Follow the journal's author instructions and the AI-use questions in its submission system. A dedicated AI declaration, acknowledgements or methods section may be requested depending on the journal and the task. Do not assume that one location works for every journal. Research-related AI use may also need methodological detail sufficient to evaluate the work; a short writing declaration does not replace that description.
Keep the manuscript statement consistent with the submission form and figure captions. If the requirements conflict or remain unclear, ask the journal before submitting. If no AI was used, state that only if requested and accurate for the whole submission.
A practical workflow for a LaTeX paper
Before an AI edit, save the working source and specify what may change. “Edit the abstract's English; preserve all numbers, citations and equations” is easier to review than a request to improve the entire paper. Afterward, compare the source, compile it, inspect the PDF and update the AI-use record. A successful build checks syntax; it cannot confirm that a scientific claim is correct.
For an EasyLaTeX project, use the safe-edit and rollback workflow and the guide to preserving meaning during AI polishing. Keep the disclosure record with your author-controlled project materials and prepare the final wording from the work actually performed.
Common questions
Does disclosure guarantee acceptance?
No. Springer Nature's guidance says declaration does not negatively influence editorial decisions. It does not promise acceptance or waive the journal's scientific and editorial requirements.
Can an AI tool be a co-author?
The linked guidance lists AI authorship as not permitted. Human authors remain accountable for the submitted work.
Must every prompt appear in the paper?
The cited manuscript-preparation page does not prescribe publishing every prompt for every writing task. Describe the purpose and extent of use accurately, retain relevant records, and provide any additional materials requested by the journal or needed to evaluate the methods.