AI did not enter scholarly publishing through policy or deliberate design. It arrived through use, becoming part of the everyday work of writing, reviewing, and editorial assessment. What stands out is not the speed of adoption, but its unevenness.
Across the system, AI is now routine. Reviewers use it to summarize manuscripts and shape reports. Authors rely on it to improve clarity and presentation. Editors are beginning to draw on AI-supported signals in triage. None of this was coordinated. It happened because the tools were available and the pressures were already there.
The scale is now clear. More than half of peer reviewers report using AI in practice. What was experimental has become baseline. What has not kept pace is alignment, and that gap is beginning to show.
AI is being used differently across authors, reviewers, and editors, often within the same journal. One reviewer may rely on it heavily, another sparingly, and a third not at all. Each approach may be acceptable, but together they introduce differences that are no longer easy to compare.
The same pattern appears in submissions. Manuscripts now arrive with varying degrees of AI involvement. This often improves readability, but the extent of that assistance differs, and it is not always visible. Editors are therefore assessing outputs that look consistent but are produced in different ways.
This is not misuse, nor is it an argument against AI. Used well, it can reduce workload and improve clarity. It is already proving useful in screening submissions and supporting reviewer selection. The issue is not use. It is alignment.
Scholarly publishing depends on distributed judgment. Reviewers assess different manuscripts, and editors make independent decisions. What holds the system together is the expectation that decisions remain comparable. That expectation weakens when the processes behind those decisions begin to diverge.
AI introduces a new variable that is applied inconsistently and often without visibility. Two reviews may look similar yet reflect different levels of assistance. Two manuscripts may meet the same expectations but be shaped in different ways. These differences do not always surface, but they accumulate.
The system continues to function, but its footing shifts. Comparability becomes harder to judge. Differences in process become harder to interpret. Confidence in decisions begins to thin, even if quietly.
This is why the conversation is moving. Early responses focused on policy, on what is allowed and what must be disclosed. Those questions remain important, but they do not address how work is actually being done.
Attention is now turning to practice. Guidance is becoming more specific, tied to roles and embedded in workflows. The aim is not to restrict AI, but to bring some coherence to its use. Adoption has already happened. The question is whether that use can be aligned in a way that preserves comparability. Without that, differences in process begin to shape outcomes in ways that are difficult to see and harder to correct.
Alignment does not require identical behavior. Scholarly publishing has never worked that way. It does require that variation stays within a range that allows decisions to be understood as consistent.
If AI continues to be used differently each time, the system absorbs that variation without fully accounting for it. If its use becomes more coherent, it can begin to support the consistency the system depends on.
This is not a technical problem. The tools already exist. It is an operational one, shaped by workflows, expectations, and how judgment is supported. AI is now part of scholarly publishing. That is settled. What remains unresolved is how its use will settle into the system, and whether that settlement will strengthen or strain it. Know more
Knowledgespeak Editorial Team
