Most publishers require authors to disclose the use of artificial intelligence. The wording differs from journal to journal, but the message is the same. If AI tools were used to generate text, improve language, create images or otherwise assist in preparing a manuscript, authors are expected to declare this.
The rationale is usually framed around transparency. Readers, reviewers and editors should know when AI has contributed to the production of scholarly work. Whether people agree with every aspect of these policies or not, the underlying principle is understandable. If AI has played a role in creating the paper, that information should become part of the scholarly record and, importantly, attached to the publication itself.
However, there is another side to this discussion that receives far less attention. While publishers are requiring authors to declare their use of AI, many journals and publishers are, I suspect, using AI within their own workflows. That raises an important question. If authors are expected to disclose AI use on each individual paper, should publishers be expected to do the same?
AI is already part of publishing workflows
Artificial intelligence is no longer confined to researchers experimenting with generative AI tools. It is increasingly being integrated directly into the infrastructure of scholarly publishing. In many cases, this is happening quietly and largely outside public discussion.
Some uses are relatively administrative. AI could help check formatting requirements, detect plagiarism, identify duplicate submissions or flag manipulated images. Other uses are potentially more significant. AI tools may assist with manuscript triage, assess whether a paper appears suitable for a journal’s scope, recommend peer reviewers, summarise reviewer reports or help editors prioritise submissions for further consideration.
Publishers may be using AI-generated summaries, graphical abstracts and promotional content linked to published articles. In some cases, AI systems may be positioned as tools that can help manage the growing pressure on editorial teams caused by increasing submission volumes and overstretched peer review systems. Bear in mind that reviewers are largely banned from using AI to peer review papers, with the reason often given that it would expose unpublished articles to the large language model being used, yet publishers may be doing this anyway.
The commercial and operational incentives are obvious. Scholarly publishing now operates at enormous scale. Large publishers process millions of submissions annually, while editors and reviewers face increasing workloads. AI can provide an efficient way to reduce costs and speed up the peer review process. It is entirely understandable why publishers are interested in integrating these technologies into their workflows.
The issue is not whether publishers should use AI. Widespread adoption is probably inevitable. The more important question is whether readers, authors and reviewers should be informed when AI has been used in the editorial/publication process?
A generic statement is not equivalent to disclosure
Some publishers already acknowledge the use of AI within policy documents or statements published on their websites. These statements may indicate that AI tools are used to support editorial operations, manuscript assessment or production workflows.
At first glance, that might appear sufficient. However, there is a significant difference between a general policy statement and a paper-level disclosure.
Authors are generally not permitted to satisfy disclosure requirements through a generic statement on a personal website. A researcher cannot simply place a sentence on their homepage saying:
“I may use AI tools to support my academic writing.”
Similarly, universities cannot satisfy journal disclosure requirements by placing a broad statement somewhere on their institutional website saying that some of their researchers may use AI during manuscript preparation.
That is not how disclosure currently works for authors. Journals increasingly require AI declarations to appear within the individual paper itself, even stating that AI has not been used. The disclosure becomes attached directly to the publication and therefore becomes part of the scholarly record. Readers do not need to search external websites or institutional policies to discover that AI was used. The information is part of the published article.
If we apply the same principle consistently, then a publisher-level statement saying:
“Our journals may use AI tools within editorial workflows”
… is not really equivalent to what authors are being asked to do.
The logical comparison would require disclosure on each individual paper where AI systems contributed to editorial assessment, reviewer selection, manuscript triage or production processes. Otherwise, we risk creating two entirely different standards of transparency. Authors must disclose AI use at paper level, while publishers can disclose theirs through broad policy statements that many readers will never see.
When does AI become part of editorial decision-making?
This discussion becomes more complicated because not every use of AI carries the same significance. Most people would probably accept that routine automation has existed within publishing workflows for many years. Spell checking, formatting validation and plagiarism screening are already deeply embedded within modern submission systems.
The more difficult question concerns the point at which AI begins influencing editorial judgement.
Suppose an AI system contributes to the initial assessment of whether a paper is suitable for peer review. Suppose it identifies “quality indicators”, “novelty signals” or “scope fit” before an editor fully evaluates the submission. Suppose reviewer recommendations are partly generated through AI systems trained on historical editorial data and citation networks.
Even if the editor retains final authority, the workflow itself may already have been shaped by algorithmic systems that remain invisible to authors and readers.
That matters because editorial decisions are not trivial administrative outcomes. Publication decisions influence careers, promotions, grant success, institutional rankings and professional reputations. A rejected paper will never be part of the scholarly archive (for that journal, at least), while an accepted paper may influence policy, healthcare, engineering or public debate.
If AI systems are increasingly contributing to those decisions, even indirectly, there is a legitimate argument that this should be declared on the article on which AI was used.
The accountability problem
There is also a broader governance issue. When authors use AI irresponsibly, accountability frameworks generally exist. Papers can be corrected or retracted. Institutions can investigate misconduct. Journals can impose sanctions. Responsibility can usually be traced back to identifiable individuals.
But what happens if AI-assisted editorial workflows contribute to problematic decisions?
If a paper is unfairly rejected because of an opaque algorithmic screening process, who becomes accountable? If an AI system contributes to biased reviewer selection or systematically disadvantages certain writing styles, disciplines or regions, where does responsibility sit?
The editor may argue that the software provider designed the system. The software provider may argue that the publisher configured the workflow. The publisher may argue that editors still retain final decision-making authority.
This becomes particularly important if publishers increasingly rely on proprietary AI systems that are not transparent to authors, reviewers or institutions. A submission could theoretically pass through multiple layers of algorithmic assessment without the author ever knowing how much those systems influenced the process.
Efficiency and transparency are not always the same thing.
Transparency should not be one-sided
None of this means publishers should avoid AI altogether. In many respects, AI may help address genuine operational problems within scholarly publishing. Reviewer shortages are real, editorial workloads are increasing, and submission volumes continue to rise. Some forms of AI assistance may improve the overall system, but we have the right to know when and how Ai was used.
Authors are increasingly expected to disclose AI use, and this is published on the manuscript, as publishers argue that readers deserve transparency. Yet publishers themselves may only disclose their own AI use through broad policy statements buried within websites or submission guidelines; if declared at all.
That creates an uneven standard. One side is expected to provide paper-level disclosure that becomes part of the permanent scholarly record, while the other may rely on generic institutional statements that remain detached from individual publications.
Academic publishing depends heavily on trust. Readers trust that papers were evaluated fairly. Authors trust that editorial processes are robust and unbiased. Institutions trust that journals maintain appropriate standards of quality and integrity.
Final thoughts
Artificial intelligence is likely to become deeply embedded within scholarly publishing over the coming years. The scale and complexity of modern publishing make that difficult to avoid. The question is no longer whether publishers will use AI. In many cases, they already are.
The more important issue is whether the use of AI within editorial and publication workflows should be visible within the scholarly record itself.
Authors are increasingly required to declare AI use, which is published as part of the article rather than through generic institutional or personal statements declared on a web site. If transparency at paper level is considered necessary for authors, it is reasonable to ask whether the same principle should eventually apply to publishers as well.
Scholarly publishing has always depended upon transparency, accountability and trust. As AI becomes more deeply integrated into publishing workflows, those principles may become more important than ever.
About the author
Graham Kendall is Acting Vice-Chancellor of GlobalNxt University, Malaysia, and Emeritus Professor at the University of Nottingham.
He writes regularly on research integrity, governance, higher education leadership and the changing dynamics of academic publishing.