Journal Publication Pro All articles
Research Integrity

Where Authors Draw the Line: Mapping Generative AI Disclosure Rules Across US Journals in 2026

Journal Publication Pro
Where Authors Draw the Line: Mapping Generative AI Disclosure Rules Across US Journals in 2026

Photo: Walters Art Museum Illuminated Manuscripts, CC0, via Wikimedia Commons

A corresponding author finishes a manuscript, runs it through a language model for one last grammar pass, uses a generation tool to render a conceptual figure, then pauses over the submission form. There is a field labeled, vaguely, something like "AI use." The honest answer is that whether any of this must be declared depends on the journal, the editor, and sometimes the reviewer who happens to notice.

Three years after generative tools became routine fixtures in academic workflows, disclosure norms in US journal publishing remain uneven. Some publishers require a dedicated statement in every submission. Others ask only for a yes-or-no checkbox. A smaller number still say nothing at all. For scholars, that inconsistency is not a bureaucratic inconvenience; it is a compliance risk that can surface as a desk rejection, a correction notice, or an awkward post-publication exchange.

Four Places AI Enters a Manuscript

Most policy confusion stems from treating AI use as a single category. In practice it appears in at least four distinct places, and editors increasingly evaluate each one separately.

Drafting and writing support. Outlining, drafting, summarizing literature, or paraphrasing. This is the most heavily policed category, largely because it touches originality.

Analysis and code. Using models to generate statistical scripts, clean data, or assist with qualitative coding. Journals care less about the tool and more about whether the analysis remains reproducible and verifiable.

Figures and images. Generative image tools are the sharpest edge of the issue. Many US journals now prohibit AI-generated or AI-altered figures except where the tool itself is the object of study, and require any permitted manipulation to be documented.

Language editing and translation. Grammar checkers and translation tools occupy a gray zone. Policies that regulate drafting often exempt light editing, but the boundary is rarely defined with precision.

What the Policies Actually Converge On

Despite the variation, a handful of principles now command broad agreement across major publishers, and the recommendations issued by bodies such as the International Committee of Medical Journal Editors and the Committee on Publication Ethics have pushed the field toward a common baseline.

First, a generative tool cannot be credited as an author. Authorship requires accountability, and a model cannot consent to or defend a claim. Second, authors remain fully responsible for everything in the manuscript, including text, references, and figures produced with AI assistance. Third, disclosure should identify the tool, its version where relevant, and the specific purpose it served. Fourth, fabricated references and AI-hallucinated citations are treated as an authorial failure, not a technical glitch.

Where journals still diverge is on placement and granularity: some want the statement in the methods section, others in acknowledgments, others in a dedicated heading at the end of the manuscript, and still others in the cover letter only. The safest approach is to satisfy the strictest plausible reading and place the disclosure where a reader would look.

Drafting a Statement That Holds Up

A defensible AI statement has four elements and fits in a short paragraph. It names the tool and version, states the purpose, describes the extent of use, and closes with an accountability line confirming that the authors reviewed and take responsibility for the output.

A workable model reads: "The authors used [tool, version] to [purpose, e.g., improve readability of the discussion section / generate initial code for the sensitivity analysis]. Output was reviewed, verified, and revised by the authors, who take full responsibility for the content of this manuscript. No AI tool was used to generate figures or to produce references."

Note what that template does. It is specific enough to be verifiable, narrow enough to be credible, and explicit about what was not done. Vague statements such as "AI was used to assist with the manuscript" invite skepticism precisely because they cannot be checked.

The Inconsistencies That Trigger Scrutiny

Most AI-related problems that reach editorial offices are not cases of deliberate concealment. They are mismatches. A cover letter that mentions a tool while the manuscript does not. A methods section that describes AI-assisted coding while a data availability statement implies the analysis was fully manual. A figure whose metadata or visual signature suggests generation despite a statement of no such use.

Three patterns draw particular attention during screening. The first is a disclosure that appears after a reviewer raises a question, which reads as reactive rather than forthright. The second is a statement that covers writing but is silent on figures, the category editors watch most closely. The third is a reference list containing sources that do not exist, a familiar hallmark of unedited model output.

Journals have also grown more attentive to the difference between AI-assisted and AI-generated. Assisting with language is treated differently from generating substantive content, and that distinction should be reflected in how the statement is worded.

A Practical Pre-Submission Routine

Before uploading, work through five quick checks. Read the target journal's policy in full rather than relying on the submission portal's checkbox. Inventory every tool used, including editors embedded in reference managers and word processors. Confirm the manuscript, cover letter, and any data or code statements agree with one another. Verify every reference manually. Finally, keep a dated internal record of which tool was used on which section, so that if an editor asks for detail, the answer is already documented rather than reconstructed.

Disclosure is not an admission of shortcut. In 2026 it is a normal part of the methods record, and the authors who handle it plainly spend less time defending their work and more time seeing it cited.

All Articles

Related Articles

The Files Behind the Paper: Why Supplementary Materials Now Decide Peer Review Outcomes

The Files Behind the Paper: Why Supplementary Materials Now Decide Peer Review Outcomes

Sign Before You Write: The Case for Formal Coauthor Agreements in Academic Research

Sign Before You Write: The Case for Formal Coauthor Agreements in Academic Research

Anatomy of a Trap: Recognizing the Sophisticated Deception Tactics Predatory Journals Use to Recruit Scholars

Anatomy of a Trap: Recognizing the Sophisticated Deception Tactics Predatory Journals Use to Recruit Scholars