Sustainability AI Policy: ChatGPT and Generative AI Disclosure Rules for Sustainability Authors

Start with a copy of your manuscript and its analysis files. Ask each contributor to complete one row per distinct AI-assisted task. Split a mixed session into separate tasks when the output moved from language editing into analysis or interpretation.

In our pre-submission worksheet, a blank verification field is a hold signal: reopen the relevant work rather than write a stronger assurance. This is an original editorial framework, not an observed customer cohort. Completing it does not prove research integrity or guarantee acceptance.

For a multi-author project, nominate one person to reconcile the rows. If one collaborator says the tool only improved English while another adopted an AI-generated interpretation, the submission needs a description of both tasks. Avoid reducing a mixed workflow to its least consequential component.

Keep private logs in an access-controlled project location. A public disclosure can describe the method without exposing confidential participant text or private prompts. If your institution restricts the underlying processing, adding a disclosure afterwards does not resolve that restriction.

The following case is fictional. It shows how the worksheet changes a submission decision; it is not a report of Manusights customer outcomes or Sustainability editorial decisions.

A team compares two packaging systems using a spreadsheet of material quantities and emission factors. A language model suggests Python code to join the tables and calculate emissions per delivered package. The same tool later edits the Discussion. The draft disclosure says only: “AI helped improve the English.”

That statement misses a method that affected the numbers. The team's first task is to identify which generated code was retained. The second is to verify the join, unit conversions, allocation choices and resulting totals against an independently calculated example.

Suppose their test finds that an emission factor is expressed per tonne while the quantity column uses kilograms. The code has multiplied the raw values without conversion. The remedy is to correct the calculation, rerun every affected result, and revisit the conclusions. Disclosure cannot make the original numbers valid.

Now inspect the Discussion edit. The original says the estimate applies to the sampled supply chain. The edited version calls the packaging “universally sustainable.” Reject that expansion unless separate evidence supports it. Restore the geographic and system boundaries even if the new sentence reads smoothly.

The completed row should connect the code to the Methods section, the software identity to Acknowledgments, and the declaration to the submission form. It should also identify the reproducible analysis files and the checks actually performed.

Our judgment is that the unit test matters more than a polished disclosure paragraph. If the corrected totals reverse the packaging comparison, the title, abstract, figures and recommendations must be reconsidered as well. If the result survives, report the supported finding with its original limitations. Neither outcome establishes broader environmental superiority.

For qualitative sustainability research, apply the same principle differently: inspect whether an AI-generated theme is supported by the interview material and whether participant permissions allowed the processing. Do not repurpose the emissions-code checks as a qualitative validation method.

Check whether your AI disclosure has the required specificity once the scientific checks are complete. A manuscript review can flag unclear reporting; it cannot validate unseen datasets or certify an institution's data-processing permissions.

The MDPI AI policy requires substantive uses to be declared at submission, explained in Methods and accompanied by product details in Acknowledgments. A cover-letter sentence alone is insufficient. The journal's Acknowledgments instructions also call for a statement identifying the use and human responsibility.

Use these original drafting prompts only when they truthfully describe your work. Replace the bracketed fields; never invent a version or claim a check you did not perform.

Methods prompt

We used [tool and version] to assist with [analysis task] in [manuscript section]. [Contributor role] checked [validation steps], corrected [issues found], and retained [identified output]. The resulting [code or materials] can be inspected at [repository location].

Acknowledgments prompt

[Tool and version] supported [task] in [sections]. We verified [checks], revised the output, and accept responsibility for this manuscript.

Completed fictional example:

The authors used ChatGPT to draft the initial table-join code for the packaging analysis. They reviewed the code, corrected its unit conversion and tested the resulting totals against a hand calculation. The authors take responsibility for the analysis. This teaching example omits version information; a real submission must supply its actual product details.

The Methods account answers how the assistance affected the research. The Acknowledgments statement identifies the assistance and responsibility. Ensure neither contradicts the actual workflow. Do not copy the worked example's validation language into a paper that has not performed those checks.

For a review or other article type without Materials and Methods, ask the editorial office where it wants the methodological account. Do not silently move the whole disclosure into a cover letter or assume a different publisher's preferred section applies.

Check whether your AI disclosure is in the right section, then reconcile the text with the declarations entered in the MDPI SuSy submission system. Keep your submitted manuscript and submission record together.

MDPI's policy addresses generative graphics alongside other substantive uses. It also applies separate image-integrity and intellectual-property requirements. The journal instructions prohibit image manipulation that misrepresents the underlying information.

Our figure checklist separates three situations. A chart rendered from verified measurements needs traceable data and a reproducible transformation. A conceptual illustration needs accurate scientific meaning and clear separation from observations. An invented image presented as an observed result is a research-integrity problem.

For each figure, write down its source, transformations and intended evidential role. Check every label and symbol against the study. Confirm rights for any third-party elements. Where acceptability remains uncertain, ask the journal before submitting the graphic. Naming the generator is not sufficient evidence of permission or accuracy.

Avoid a blanket “all AI figures are prohibited” rule. Equally, do not interpret a disclosure provision as blanket approval for any generated image. The practical decision depends on what the image represents, how it was produced and whether it satisfies the applicable integrity requirements.

A disclosure prepared for Sustainability should be reviewed again if you submit elsewhere. This comparison was checked in the publishers' public pages on September 12, 2026; it is a transfer checklist, not a complete substitute for the destination journal's instructions.

The worksheet still helps because it records what happened. The destination policy determines how that record must be reported. Do not replace an accurate account of analysis assistance with a language-editing statement during a transfer.

Under the MDPI reviewer AI rules, reviewers must not generate review reports with GenAI or upload confidential manuscripts and related material. Limited wording assistance may be acceptable with disclosure and confidentiality preserved.

This role distinction matters when an author also reviews for Sustainability. A workflow used on one's own manuscript is not automatically permitted on someone else's confidential submission. Broader ICMJE recommendations and COPE's AI authorship position provide context; the exact MDPI policy controls this submission.

If a tool both edits wording and adds interpretation, separate the two effects. Keep the editing exemption scoped to superficial changes. Ask the journal when a material use remains difficult to classify, describing the actual task rather than just the product name.

If a disclosure was omitted, correct the record with the editorial office. Do not predict an automatic rejection, fixed delay or acceptance probability. Consequences depend on the facts and the applicable ethics process. A disclosure omission and fabricated research evidence are different problems, even when they occur together.

After publication, substantive errors may require a correction or other editorial action under MDPI's publication-ethics procedures. Contact the journal with an accurate account. Do not quietly alter a public method description while leaving affected results unexamined.