Science Of The Total Environment AI Policy: ChatGPT and Generative AI Disclosure Rules for STOTEN Authors

Elsevier provides a standard declaration format. For a STOTEN environmental-science paper, make the statement concrete enough that an editor can see what the tool did and what it did not do.

"During the preparation of this work, the authors used ChatGPT-4o to improve language clarity in the Introduction and Discussion. After using this tool, the authors reviewed and edited the text, verified all references and environmental claims, and take full responsibility for the content of the published article. Generative AI was not used to generate data, alter figures, conduct statistical analysis, or write the final interpretation."

For AI-assisted code or modeling, do not hide the method detail in a generic declaration:

"Claude Sonnet 4 was used to draft initial Python code for the sensitivity-analysis workflow. The authors rewrote and validated the code, verified all outputs against the raw dataset, and archived the final script with the analysis files. The model was not used to fabricate data or select results."

Weak declarations create avoidable risk:

• "AI tools were used in manuscript preparation." This names neither the tool nor the use.
• "ChatGPT helped write this paper." This does not say what was reviewed or who takes responsibility.
• "AI assistance is acknowledged in the cover letter." STOTEN's guide points authors to a manuscript declaration section.
• "No AI was used" when AI-assisted translation, literature synthesis, code drafting, or figure drafting actually shaped the manuscript.

STOTEN's guide includes a separate "Generative AI and Figures, images and artwork" section. It says generative AI or AI-assisted tools are not permitted to create or alter submitted manuscript images, and it gives a narrow exception when the AI tool is part of the research design or methods and is described reproducibly in Methods.

Elsevier's newer publisher-wide journal policy uses more detailed image categories. It allows some explanatory images only with transparency, does not allow AI to fabricate or alter primary research images, and treats data visualizations as acceptable only when they are faithfully derived from underlying data and reproducible methods. For STOTEN, the practical reading is conservative: if the visual carries environmental evidence, treat it as research output, not decoration.

This is where STOTEN differs from a generic AI-writing page. Recent STOTEN papers span evidence types such as neonatal cadmium exposure data (10.1016/j.scitotenv.2025.179011), satellite and land-use regression estimates for PM2.5 ionic composition (10.1016/j.scitotenv.2025.180050), monsoon and sandstorm modeling (10.1016/j.scitotenv.2024.177874), and natural seaweed-bed image processing (10.1016/j.scitotenv.2025.179296). In those manuscript types, an AI-altered map, plot, segmentation image, or workflow diagram can change what the reader thinks the evidence shows.

Verify the current Editor-in-Chief on the journal's editorial-team page before quoting any name in a cover letter, because editor assignments and board pages can change.

Before submission, separate visuals into three buckets:

In our pre-submission review work for Science of The Total Environment and adjacent environmental journals, the AI-policy issue is rarely that authors used a tool at all. The more common problem is that the disclosure is too thin for the manuscript's evidence burden.

The pattern looks like this: the paper has strong environmental data, but the authors used AI to smooth the introduction, draft limitations, translate sections, or produce starter code for a model. The declaration then says only "AI tools were used." For STOTEN, that is not enough context. A reader needs to know whether the tool touched the claims, the citations, the Methods, the figures, or only surface language.

The highest-risk STOTEN submissions are not the clean language-editing cases. They are papers with AI-assisted geospatial classification, exposure modeling, source-apportionment code, image processing, or automated literature synthesis, because those uses can affect the scientific record. If your manuscript includes those workflows, treat the AI statement as part of reproducibility, not as a compliance footnote.

Common STOTEN AI-policy failure patterns

Generic disclosure on a methods-heavy paper. A manuscript reports machine-learning exposure estimates, pollutant-source attribution, or remote-sensing classification, but the AI declaration only says that an AI tool improved language. If AI helped draft code, summarize literature, choose model wording, or write the limitations, the declaration needs to separate those uses from surface editing.

Image workflow hidden behind a polished graphical abstract. STOTEN authors often prepare maps, conceptual pathways, sampling-site diagrams, and graphical abstracts late in the submission package. When generative AI helped create or alter those visuals, the risk is not cosmetic. The visual can imply source locations, exposure pathways, or mechanistic relationships that are not in the data.

Translation or paraphrase that changes environmental claims. Some STOTEN papers start in another language and are then rewritten with AI. The problem is not translation itself. The problem is when terms such as "risk," "hazard," "exposure," "remediation efficiency," or "policy implication" are strengthened during rewriting without author verification.

AI-assisted citation discovery without manual verification. Environmental review sections are prone to citation drift. If AI suggested sources, every citation still needs to be checked against the paper's actual claim, especially when the manuscript cites contaminant limits, health-risk thresholds, or regulatory guidance.

A missing declaration can create several outcomes: a submission query, rejection before review, a correction request after acceptance, or a publication-ethics review if the undisclosed use affected data, images, references, or conclusions. Do not assume STOTEN will treat all AI use the same way.

The ambiguous cases are the ones authors mishandle:

• AI translation of author-written text: disclose if the tool materially rewrote the manuscript, and make sure a bilingual domain expert checked the output.
• Reference discovery with AI: disclose if the tool selected or summarized sources, and manually verify every citation.
• Code generation for analysis: disclose in Methods if it shaped the research workflow, and archive the final validated code when possible.
• Figure polishing: avoid generative AI for submitted manuscript visuals unless the current STOTEN guide and Elsevier policy clearly allow the exact use.
• Reviewer-side AI use: STOTEN's guide currently says the journal does not allow ChatGPT or similar services for reviewers or editors in peer review and manuscript evaluation, while Elsevier's publisher page has broader reviewer guidance for private supportive tools. Authors should follow the journal guide and not infer reviewer behavior from the publisher summary.