Elsevier Now Requires a Separate AI Declaration, and Bars AI From Data Images

Elsevier’s policy separates three things that are often run together: AI used to help write the paper, AI used in the research itself, and AI used to make pictures. Each has its own rule and its own place for the disclosure.

Writing. Use for manuscript preparation is disclosed “in a separate AI declaration statement”, which the policy says appears in the published article. The declaration names the tool, says what it was for, and says how the author oversaw it. Basic grammar, spelling and punctuation checks need no declaration, and neither does assistive technology used for accessibility.

Research. AI used as part of the research goes in the Methods, like any other method.

Images. This is where the policy is strictest. Creating figures, images and artwork with AI is “only permitted in specific circumstances”. AI “must not be used to create or alter images that represent primary observed or experimental data” — microscopy, blots, clinical images. General-purpose generative AI image tools “must not be used to create graphical abstracts”. Where an explanatory image is permitted, the tool, version and use go in the caption as well as in the article-level statement.

Reviewers and editors. “Reviewers should not upload a submitted manuscript or any part of it into an AI tool”. The same applies to editors. This is a confidentiality rule, not a quality one: an unpublished manuscript pasted into a public tool has left the review process.

The image rule is the one that reveals the thinking. A declaration is about transparency and can be fixed after the fact; an altered data image is evidence of something that did not happen, and no declaration repairs it. So Elsevier treats writing assistance as disclosable and data imagery as prohibited.

The reviewer rule follows the same logic applied to confidentiality. A manuscript under review is the author’s unpublished work, held in confidence. Uploading it to a third-party service is a disclosure to that service, whatever the reviewer intended, and it cannot be undone.

The carve-outs for grammar checkers and accessibility tools matter too: a policy that required a declaration for a spellchecker would be ignored, and one that discouraged accessibility tooling would penalise the authors least able to absorb it.

How it reaches a specific journal. Elsevier publishes the policy at publisher level, and individual journals implement it through their own submission systems — typically a declaration field at submission plus the statement in the manuscript. The Lancet titles and the Cell Press journals are Elsevier imprints and sit under it.

Where a journal’s own instructions say something stricter, the stricter rule governs for that journal. The publisher policy is a floor, not a ceiling.