On the Peer-Review Crisis

Peer review decides which scientific claims enter the permanent record, which researchers receive recognition and which ideas disappear before the public ever sees them. Yet the system responsible for these decisions is increasingly overloaded, difficult to audit and dependent on experts who are usually unpaid, time-poor and rarely trained for the task.

This is the peer-review crisis. It is not simply a complaint about slow journals or unfair reviewers. It is a structural failure in the mechanism science uses to evaluate itself.

In a widely cited critique, former medical-journal editor Richard Smith argued that peer review was supported by “faith in its effects, rather than facts.” His assessment described a process that can be slow, inconsistent, biased and poor at detecting fraud, despite remaining central to scientific publishing. Read Smith’s analysis of the peer-review system.

The Uncomfortable Truth: Peer Review Runs on Invisible Labour

Modern science produces more manuscripts, more specialist methods and more interdisciplinary research than the traditional review system was built to handle. Every new submission requires editors to locate qualified experts who are willing to interrupt their own research, teaching and administrative work to evaluate somebody else’s paper.

Most of this labour is performed without direct payment. Reviewing may support professional reputation or disciplinary service, but it competes with activities that universities formally measure: publications, grants, teaching and institutional responsibilities.

The result is a predictable bottleneck. Editors send repeated invitations. Researchers decline because they are overcommitted. Reviews arrive late, or they are completed under severe time pressure.

An international study involving more than 4,000 researchers found that scientists continued to value peer review and generally believed that it improved published papers. The same study, however, documented concerns about workload, fraud detection and the need for reviewer training. More than half of the reviewers represented in the peer-review crisis synthesis spent fewer than five hours evaluating a manuscript. Explore the international study of researcher attitudes toward peer review.

The problem is therefore not that scientists have stopped caring about quality. The problem is that the system asks them to protect quality without providing sufficient time, recognition or infrastructure.

The Gatekeepers Are Rarely Trained to Guard the Gate

Peer reviewing is a specialised form of scholarly judgment. A reviewer may need to assess research design, statistical analysis, reporting standards, ethical compliance, theoretical coherence, novelty and reproducibility. Nevertheless, journals do not normally require reviewers to demonstrate formal competence before evaluating manuscripts.

A 2023 international survey found that 39% of respondents had never received peer-review training, while 35.8% had trained themselves by reading academic literature. More than half recalled feeling at least slightly unprepared when conducting their first review. Read the international survey on peer-review training.

This creates a strange contradiction. Authors are expected to follow detailed methodological and reporting standards, but the people assessing their work may have learned reviewing through observation, imitation and trial and error.

Experienced researchers can become excellent reviewers through practice. Experience alone, however, does not create a shared standard. Two competent reviewers may examine the same manuscript and focus on entirely different questions, apply different thresholds and reach conflicting recommendations.

The System Is Secretive by Design

Traditional peer review usually happens behind closed doors. Review reports remain confidential. Reviewer identities may be hidden. Editorial discussions are rarely published. Rejected manuscripts frequently leave no accessible review history.

Confidentiality can protect reviewers from retaliation and encourage direct criticism. But secrecy also makes the process difficult to study, compare or challenge.

Open peer review has been proposed as one response, but even that term lacks a single agreed meaning. A systematic review identified 122 definitions and 22 configurations of open peer review, including open identities, published reports, public participation, author-reviewer interaction and post-publication commentary. Read the systematic review of open peer-review models.

The deeper problem is not merely that reviews are hidden. It is that journals use different procedures, instructions, scoring systems and decision rules. This variation makes it difficult to determine why one manuscript was accepted while another received rejection for apparently similar weaknesses.

Questionable Peer-Review Practices Threaten the Scientific Record

The crisis also includes practices that directly compromise publication integrity. These include manipulated reviewer suggestions, fabricated reviewer identities, reciprocal reviewing arrangements, inappropriate alteration of reports and reviews written by people with undisclosed conflicts of interest.

A 2024 paper grouped these behaviours under the term questionable peer-review practices. Its proposed responses included stronger editorial oversight, reviewer training, open review, crowd review and wider adoption of Registered Reports. Read “Questionable Peer Review Practices”.

Fake review activity has also contributed to mass retractions and investigations. Paper mills and organised review manipulation can exploit editorial systems by submitting false reviewer details or controlling supposedly independent reports. Read Nature’s reporting on the scale of fake peer review.

These failures matter because publication creates legitimacy. Once an unreliable paper enters a respected journal, it may influence later experiments, clinical guidance, public policy, funding decisions and media coverage long before its weaknesses are discovered.

Peer Review Can Be Too Slow—and Dangerously Fast

Slow peer review delays the circulation of knowledge. Researchers may wait months for a first decision, only to restart the process at another journal after rejection. The delay is especially costly in rapidly developing fields, public-health emergencies and research with immediate environmental or clinical relevance.

Yet speed creates its own danger.

During the COVID-19 pandemic, medical journals dramatically accelerated publication. One study found that average submission-to-publication time for coronavirus-related papers fell from approximately 98 days to 50 days. Read the study of accelerated pandemic publishing.

This demonstrated that the publishing system could move faster when placed under pressure. It also exposed the trade-off between urgency and scrutiny. A review completed too slowly can obstruct scientific progress. A review completed too quickly can overlook methodological errors, unsupported claims or unreliable evidence.

The solution is not simply faster review. It is faster access to structured scrutiny without reducing the time available for consequential human judgment.

Interdisciplinary Science Is Punished for Crossing Boundaries

Some of the research most capable of addressing complex global problems also creates the greatest difficulty for conventional peer review.

An interdisciplinary manuscript may combine unfamiliar theories, methods, datasets and vocabularies. An editor may struggle to find one reviewer qualified to evaluate the entire work. Multiple specialists may be needed, each assessing only part of the paper.

A 2024 study of publication delays found that interdisciplinary research faces distinctive disadvantages during journal evaluation. Reviewer selection becomes harder, disciplinary distance increases and the manuscript may be judged through standards designed for narrower fields. Read the study of delays in publishing interdisciplinary research.

This creates a damaging incentive. Researchers are encouraged to solve problems across disciplinary boundaries, but the publication system may penalise them for doing so.

Bias Does Not Need to Be Deliberate to Shape Science

Peer-review bias is not limited to openly discriminatory behaviour. Decisions may be affected by institutional prestige, country, language, gender, professional networks, theoretical preferences or assumptions about what important science should look like.

Research examining manuscript outcomes has reported disadvantages for authors affiliated with parts of Asia, countries where English is not the primary language and countries with lower Human Development Index values. Review the evidence on barriers affecting historically excluded groups.

Another large analysis found that women and researchers outside North America and Europe were underrepresented among authors, editors and reviewers. It also identified patterns of author-reviewer similarity that could influence review outcomes. Read the study of author-reviewer homophily.

No review system can eliminate every difference in scientific judgment. But a system that hides its reasoning makes it harder to distinguish legitimate disagreement from systematic disadvantage.

The System That Audits Science Is Itself Difficult to Audit

Peer review is expected to expose weaknesses in evidence. Yet researchers studying peer review often cannot access the evidence needed to evaluate the process itself.

Public datasets frequently contain only final papers and isolated review comments. They may omit the original submission, revised manuscripts, editorial decisions, meta-reviews or author rebuttals. Without these records, it is difficult to determine what changed during review, which comments influenced revision and how final decisions were reached.

The Multidisciplinary Open Peer Review Dataset, or MOPRD, was created partly to address this evidence deficit. It includes metadata, manuscript versions, review comments, meta-reviews, rebuttal letters and editorial decisions for 6,578 papers.

The significance of such datasets extends beyond artificial intelligence. They make it possible to study the behaviour of reviewers, the consistency of editorial decisions, the effect of criticism on manuscript revision and the differences between disciplines.

A quality-control system cannot improve reliably while most of its operations remain unavailable for inspection.

Generative AI Is Arriving in the Middle of the Crisis

Generative AI has intensified the pressure on scholarly publishing. It can help researchers draft, translate, summarise and restructure manuscripts. It can also increase submission volume, generate fabricated references and conceal low-quality or synthetic research behind fluent academic language.

At the same time, AI can assist editors and reviewers with administrative screening, reporting checks, statistical verification, evidence retrieval and structured manuscript analysis.

This creates a dangerous temptation: replacing an overloaded human process with an automated decision system.

The available evidence does not justify that step.

The 2026 xPeer benchmark compared two human reports and two xPeer simulation reports for 271 manuscripts. xPeer produced longer reports, more extracted concerns, broader category coverage, stronger manuscript targeting and more explicit revision actions. Human reviewers showed more explicit reasoning, stronger lexical connection to manuscript passages, slightly higher scientific relevance and lower repetition between paired reports.

The study therefore found different review profiles rather than a universal winner. It explicitly reserved scientific correctness, severity and editorial utility for expert adjudication. Explore the open xPeer benchmark dataset on Zenodo.

The strongest case for AI in peer review is not reviewer replacement. It is structured, reproducible and auditable support for reviewers who remain responsible for scientific judgment.

What Would a Repaired Peer-Review System Look Like?

Peer review cannot be repaired by demanding that researchers work faster. Sustainable reform must change the infrastructure and incentives surrounding the work.

Reviewer training: Researchers should receive practical instruction in methodology, ethics, bias, reporting standards and constructive criticism before conducting independent reviews.
Recognition and incentives: Reviewing should be measured and rewarded as a formal scholarly contribution rather than treated as invisible service.
Transparent procedures: Journals should publish clearer evaluation criteria, decision pathways and, where appropriate, review histories.
Structured review frameworks: Checklists, reporting standards and stable review tasks can reduce accidental omissions without removing expert discretion.
Open evidence for research: Anonymised review records should be made available for studying quality, consistency, bias and disciplinary variation.
Human-governed automation: AI tools should surface concerns, test completeness and support analysis while leaving consequential decisions to accountable humans.
Appeal and contestability: Authors should be able to challenge demonstrably incorrect, irrelevant or procedurally unfair review outcomes.

The Crisis Is Structural, Not Personal

It is easy to blame careless reviewers, demanding authors or slow editors. Most participants, however, are operating inside a system that asks a limited pool of experts to evaluate a rapidly expanding and increasingly complex research literature.

The peer-review crisis is the product of this mismatch.

Science still needs expert criticism. It still needs independent scrutiny, methodological challenge and editorial judgment. But it can no longer depend exclusively on an opaque process powered by unrecognised labour and inconsistent local practices.

The future of peer review will require trained humans, transparent evidence and carefully governed analytical systems working together. Anything less risks preserving the appearance of quality control while its foundations continue to weaken.

For organisations exploring structured, human-governed peer-review simulation, the xPeerd API Server provides programmatic access to the xPeer review engine.