审稿人池扩大 54% 同行评审却更糟：Prophy 认为问题不在供给而在注意力

In the first quarter of 2026, journals on ScholarOne Manuscripts received 33% more submissions than a year earlier. The year before, growth was 17%. Josh Dahl published the figures on The Scholarly Kitchen in May, and every peer-review panel since has quoted them, including one Prophy convened in July.

Everyone who quotes the number arrives with the same diagnosis: reviewer fatigue. More manuscripts, the same reviewers, the same hours. In other words, this is a supply problem, so the fixes are supply fixes: recruit more, recognize more, hand the mechanical checks to machines.

I believe the community has misunderstood the reviewer fatigue problem. It is not about the scarcity of reviewers. Reviewers are simply not asked about the right manuscript. I can see this in independent numbers. I can see this in how Prophy clients act.

Take one view on reviewer fatigue, and the cure is more invitations. Take the other, and the answer is better-aimed invitations.

One number, two diagnoses

The common reading: submissions at many publishers are rising 20–30% a year, while the number of reviewers, and the hours they can give to reviewing, have stayed flat.

Lucia Steele, who runs publications at AboutScience, sees a small rise in submissions and a sharp fall in engagement: “there is a huge amount of reviewers, they receive a lot of requests, and it’s difficult to actually get through to the correct one, to the best one.”

These are different diagnoses and they ask for different treatments. Take one view on reviewer fatigue, and the cure is more invitations. Take the other, and the answer is better-aimed invitations.

The panel never reconciled the two. It could not, because both are true, of different parts of the same system.

What ScholarOne’s own numbers say

While solving the supply problem, we manufactured the attention problem.

How a growing pool was spent

In the past, peer review worked because handling editors knew the field and the people in it. Matching was not a step in the process. It happened by itself, because the editor could name the three right reviewers from memory.

Publishing then scaled that model far past the point where anyone knows the pool. Submissions arrive from everywhere, in subfields no single person follows, and what replaced memory was volume: send more invitations and let the response rate sort them out.

That is where a 54% larger pool went. It was spent rather than lost, one poorly aimed invitation at a time, until the people in it stopped opening the email. Reviewer fatigue is not a symptom of too few reviewers. It is what a growing pool looks like after years of being asked about the wrong manuscripts.

I see this from both sides of the table. Prophy has matched reviewers for publishers and funding agencies since 2018, and as a working physicist I receive the invitations. The pattern is the same.

The cure is better aim

If the constraint is attention, more invitations cannot be the cure. Better-aimed ones can, and that is what Prophy is designed to address. An invitation lands when three conditions hold.

The recommendation carries its reasoning. Someone about to spend a scarce invitation wants to see why this person: what in their recent work matches the manuscript, whether they have co-authored or shared an institution with the authors, how much they have published and how recently. A name with a score is a name. A name with its evidence is a decision the editor can defend, and a short list they can stand behind, which is the only cure for the widen-the-list reflex.

The matching happens at the level of the science, not its keywords. That is what separates the first invitation from the fourth. The Prophy team built and maintain our own ontology of about 170,000 scientific concepts for that reason, curated for research assessment rather than pulled from a general index at query time.

And the invitation comes from a person. Lucia Steele’s sharpest observation was that a direct invitation from someone who knows the reviewer outperforms any automated system. I agree, and I read it as a specification: the tool prepares the ground, the editor sends the invitation, and the relationship stays human. It follows that the tool has to sit where the manuscript already is. Outside the submission system it becomes a step reserved for the hard cases, exactly where the editor’s own network has run dry.

The 33% is not going away; Josh Dahl’s own reading is that AI-assisted drafting has made submitting structurally easier. But the reviewer pool grew 54% in the same period, which should retire the idea that supply alone is the constraint. What is scarce is attention, and in the era of AI, attention is all you need.

注意转载授权 · 阅读原文