Policy Significance Statement

This editorial reflects on how we arrived here, what it means for scientific advancement, and what responsible editorial practice must look like in an era of AI-assisted knowledge generation.

1. How we got here

The crisis has structural roots. The transition to electronic publishing removed one of the most natural constraints on journal output: the physical limit of what could fit between two covers. Once that ceiling lifted, publication volumes expanded accordingly—and with them, submission volumes. Meanwhile, academic career structures doubled down on quantity. Hiring committees count publications; promotion panels weigh output. The incentive is not to write one important paper but to write many adequate ones. The result is a literature increasingly dense with incremental contributions, each requiring peer review, each making a claim on the finite and largely uncompensated time of the scholarly community.

Artificial intelligence has accelerated this dynamic considerably. The ease with which AI tools can assist in drafting, structuring, and polishing manuscripts has lowered the barrier to submission in ways that are not uniformly matched by a corresponding improvement in conceptual originality or argumentative rigor. Submission volumes have risen; however, the average quality of what arrives in editorial inboxes has not risen with them. The peer review system—already stretched—now absorbs this additional load without any structural adaptation to match it.

Distributed editorial models, introduced partly to democratize decision-making and reduce bottlenecks, have added a further complication. While they broaden participation, they also introduce inconsistency: in how reviewers interpret scope, in what threshold of rigor they apply, and in how editorial decisions are calibrated across a diverse board. For interdisciplinary journals, where submissions traverse multiple disciplinary traditions, each with their own epistemological conventions and quality benchmarks, this inconsistency is particularly consequential. A paper operating at the boundary of computational science and public policy governance requires reviewers willing to step outside their home tradition and engage seriously with an unfamiliar intellectual language. That is already a high aspiration. Asking those reviewers to do so at ever-greater volume, with ever-less institutional recognition, is asking them to do it under conditions that make meaningful engagement genuinely difficult.

The honest conclusion is this: academic publishing with human peer review, in its current form, is not a sustainable model.

2. What is actually at stake

It would be a mistake to treat this as merely an administrative inconvenience. Peer review is not incidental to scientific advancement—it is constitutive of it. The process by which claims are scrutinized, challenged, refined, and ultimately accepted or rejected by a community of peers is how knowledge is validated and how a field coheres. When that process is under strain, the consequences are epistemic, not merely operational.

This matters especially for a field like Data & Policy, which is still in the process of building the shared conceptual vocabulary, methodological norms, and institutional infrastructure that more established disciplines take for granted. High-quality peer review is not just quality assurance here—it is field-building. A reviewer who takes the time to engage seriously with a submission that bridges, say, algorithmic governance and constitutional law is doing more than assessing a manuscript: they are helping to articulate what rigor looks like at that intersection, and in doing so, they contribute to the intellectual foundations of the field itself. The erosion of conditions that make such engagement possible is therefore not just a publishing problem. It is a problem for the coherence and credibility of the field.

The entry of AI into knowledge generation adds a further dimension. AI tools can accelerate literature synthesis, surface relevant prior work, and assist in identifying structural weaknesses in arguments. These are genuine contributions to the editorial process. But scientific advancement is fundamentally about human understanding—about the capacity of a community of scholars to make sense of the world, to contest interpretations, and to build knowledge that is accountable to both evidence and reason. No AI system currently participates in that process as a principal. It can support it; it cannot substitute for it. Clear, principled guidelines on where and how AI may be used in editorial and review processes are therefore not a bureaucratic nicety—they are a condition for maintaining the integrity of the scholarly record.

3. What responsible editorial practice requires

If the current model is unsustainable, the response cannot simply be to absorb more pressure or to automate our way out. It requires deliberate structural choices. We offer three here that reflect our own evolving editorial commitments at Data & Policy.

First, create conditions that encourage quality over quantity. One practical measure worth serious consideration is limiting the number of submissions any single corresponding author may make within a defined period—for instance, one submission per year per author. The intent is not punitive but corrective: to encourage authors to submit selectively and invest the effort that genuine quality requires, rather than distributing risk across multiple incremental submissions. This is a modest intervention, but it sends a clear signal about what kind of scholarly community a journal is trying to be.

Second, invest in editorial infrastructure before peer review begins. The most consequential reform is also the most resource-intensive: dedicating serious professional attention to the initial editorial review, before any submission reaches a peer reviewer. This means a qualified human editor—someone with genuine standing in the field, an understanding of cross-disciplinary publication norms, and the judgement to assess both scholarly merit and fit—making considered decisions about which submissions are ready to be seen by reviewers and which are not. Administrative support and AI tools can and should assist in this process: screening for basic standards, flagging inconsistencies, and managing workflow. But the decision itself—the act of judgment about whether a piece of work is ready to enter scholarly conversation—must rest with a human who can be held accountable for it and who has the standing to make it credible.

Third, treat reviewer and advisory board time as what it is: a scarce and valuable resource. Every time a reviewer opens a submission that should have been desk-rejected, the journal has made an implicit statement about how it values that reviewer’s expertise and goodwill. But volume is only half the problem. An equally significant failure of editorial respect is the mismatched request—asking a reviewer to engage with a submission that sits outside their genuine area of expertise, or that connects to their work only tangentially. A targeted, well-matched review request, sent for a paper that has already cleared a genuine quality threshold, signals something important: that the editor understands who the reviewer is, what they know, and why their particular judgment matters for this particular piece of work.

Editorial respect for voluntary scholarly time is not merely a courtesy—it is a condition for maintaining the ecosystem on which peer review depends. Demonstrating that respect concretely means ensuring that what reaches reviewers has passed a genuine quality threshold; that requests are targeted, proportionate, and matched to genuine expertise; and that the contributions of advisory and editorial board members are visible, acknowledged, and not taken for granted. When these conditions are met, reviewing is what it should be: a meaningful act of scholarly participation, not an administrative burden visited upon the willing.

4. A note on AI and editorial integrity

We are aware that AI is already present in the editorial process—on the author side, and increasingly on the reviewer and editorial side too. We do not think the appropriate response is prohibition. AI-assisted screening, AI-assisted reviewer matching, and AI-assisted structural feedback all have legitimate roles to play in managing volume and improving consistency. What we insist on is that these tools operate in support of human judgment, not in place of it. The decision to accept, reject, or request revision of a submission is a scholarly act—it carries epistemic weight and community consequences that cannot be responsibly delegated to a system that does not participate in the knowledge community whose norms it would be enforcing.

Data & Policy is committed to developing and publishing clear guidelines on AI use across the editorial process—for authors, reviewers, and editors—and to being transparent about how and where these tools are used in our own operations. We invite our community to engage with us on what responsible practice should look like.