AJPS revises its AI policy: name the tool, the version, and who checked the output

Most journal AI policies ask a yes or no question: did you use a generative tool, and if so, say so. The revised policy that the American Journal of Political Science published on 11 August 2026 asks a harder one. It asks how closely a human actually checked the output.

What happened

AJPS, the flagship journal of the Midwest Political Science Association, published a revised AI policy for authors and reviewers on 11 August 2026. The policy itself carries the same update date.

The editors describe it as the outcome of months of dialogue with their editorial staff, with journal editors inside and outside political science, with their publisher Wiley, and with the scholarly community. They also say plainly that the discipline is divided: some scholars welcome AI for idea generation, theory generation, literature review, writing, and data collection and analysis, while others oppose any use beyond something as narrow as copyediting.

What authors have to declare

The policy sets a governing principle first. Authors may use AI "only as a companion to all their work," and they "must not use AI as a replacement for their own judgment." Responsibility for the accuracy of every claim, citation, reference and analysis stays with the author. AI tools cannot be listed as authors, on the grounds that they cannot meet the accountability requirement of authorship.

Disclosure is owed for any use that shaped the research or the manuscript, including generating or editing text, generating images or figures, collecting or managing data, generating or editing code, processing or analysing data, and informing research design. It takes the form of a single consolidated statement in a manuscript footnote, mirroring the statement submitted through the journal's Author Questionnaire.

Authors are asked to separate two kinds of use:

• uses that directly shaped the research or the manuscript content, such as collecting data, generating analysis code, or producing text that appears in the paper
• uses that supported the writing process, such as proofreading, formatting, citation verification, or revising individual passages

For the first category, the reporting requirement is unusually specific. Authors should name every tool, give the provider where the name does not make it obvious, state the task, and state the degree of human oversight, meaning whether AI output was reviewed, edited, or approved before it went into the manuscript. Where available they should also give the version and the approximate dates of use.

Basic spelling and grammar correction and minor formatting are exempt. The policy closes that door carefully: the exemption "does not extend to any use that involves data, analysis, or substantive content, however minor."

Non disclosure carries a stated consequence. In the journal's words, failure to disclose as required "is itself grounds for decline."

What reviewers can and cannot do

Manuscripts under review are confidential and must not be uploaded to AI tools, in full or in part. That restriction follows Wiley policy.

Reviewers may use AI to organise their thoughts and prioritise feedback, to improve the clarity and tone of a review, to translate it, and to ask general questions about research methods or statistical approaches, provided they do not reference specifics of the manuscript. Reviewers must disclose AI use for drafting assistance.

Context

This lands while a broader standard is being built. In March 2026 COPE announced that it had joined the International Science Council, the World Conferences on Research Integrity Foundation, STM and the Global Young Academy in work toward a Global Reporting Standard for AI Disclosure in Research. The initiative was a focus track at the World Conference on Research Integrity in Vancouver from 3 to 6 May 2026, aims to complete by the end of 2026, and is intended to produce what the group calls the Vancouver Standard.

Until something like that exists, a researcher submitting to several journals is reading several different policies. AJPS is explicit that its own is constrained by Wiley's, and that some of Wiley's concerns are legal ones such as copyright.

Analysis

The following is Bizarus commentary, not reporting.

The interesting move here is not the disclosure requirement. It is the request for the degree of human oversight.

Asking whether a tool was used tells an editor almost nothing about the reliability of what follows. A fabricated citation and a carefully checked one look identical on the page, and the difference between them is entirely a matter of what the author did after the output appeared. By asking authors to state whether output was reviewed, edited, or approved, AJPS is asking for a description of process rather than a description of tooling. That is the variable that actually predicts whether the fabricated reference gets caught.

It is worth being clear about the limits. Version and dates are requested only "when available." The exemption for minimal uses has a boundary that the editors themselves acknowledge is a grey area, which is why the policy tells anyone in doubt to contact the journal directly. And a disclosure regime of this kind is self reported, so it constrains the conscientious more than the careless. None of that makes it weak. It makes it a policy that has understood what it can and cannot enforce, which is more than a blanket prohibition would achieve.

The phrasing that stands out is the requirement that AI be a companion to the work rather than a replacement for the author's judgment. That is the distinction this platform exists to hold, and it is notable to see it doing load bearing work in the policy of a major journal rather than sitting in a preamble.

What to watch

Whether the Vancouver Standard, if it arrives on schedule at the end of 2026, converges on oversight reporting of this kind or settles for tool disclosure alone. The two produce very different amounts of useful information for an editor.

Sources

• American Journal of Political Science, Revised AJPS Artificial Intelligence (AI) Policy for Authors and Reviewers, 11 August 2026
• American Journal of Political Science, AJPS AI Policy, updated 11 August 2026
• COPE, Global reporting standard for AI disclosure in research, 25 March 2026