The State of Peer Review in 2026: More Transparent, More Automated, and More Stressed

The biggest structural change is not actually transparent peer review or AI. It is research integrity.

In January 2026, STM released a report saying some publishers now maintain dedicated research-integrity teams numbering more than 100 staff and screen millions of manuscript submissions annually. The same announcement said the STM Integrity Hub had 49 organizational members, COPE had 106 publisher members representing more than 14,500 journals, and United2Act had 58 organizations coordinating responses to paper mills.

That is not business-as-usual scale. It is defensive infrastructure.

What this means for authors:

• image and data screening is more normal
• identity and authorship questions are more likely to be checked
• suspicious citation patterns, reviewer suggestions, or paper-mill signals are more likely to trigger scrutiny

In other words, peer review is no longer just experts reading a manuscript. It increasingly begins with systems trying to decide whether the manuscript and its provenance can be trusted at all.

Transparency has been discussed for years. In 2025 and 2026 it moved closer to default practice in major parts of the system.

Nature announced that for new research submissions published from June 16, 2025, peer-review files would automatically accompany published research articles. The editorial notes that Nature had offered this as an opt-in option since 2020, and that Nature Communications had been doing transparent peer review since 2016.

Nature Communications itself now publishes reviewer comments and author rebuttal letters for all original research articles accepted from submissions received on or after November 1, 2022.

This matters because transparent peer review changes incentives:

• authors know rebuttals may be visible
• reviewers know their reports may become part of the public scientific record, even if anonymous
• readers get a clearer view of how claims were challenged and revised

It still does not reveal everything. Nature Communications explicitly says editorial discussions and confidential comments are not included. But compared with the old black-box model, the shift is real.

This is the most unstable area right now.

On the policy side, major publishers are drawing clear lines. Elsevier says reviewers should not use generative AI to assist in the scientific review of a paper and should not upload manuscripts or reports into AI tools because of confidentiality and quality risks. Springer Nature's current AI guidance says manuscripts should not be uploaded into generative AI tools and that editorial and peer-review assessments must be made and verified by humans. Nature Methods reiterated in February 2026 that uploading manuscripts into generative AI tools is not allowed during peer review.

On the behavior side, researchers are clearly using AI anyway.

Frontiers reported in December 2025 that 53% of reviewers in its global survey of 1,645 active researchers said they now use AI tools in peer review. Early-career adoption was even higher.

That gap between policy and behavior is one of the defining tensions of 2026.

The likely near-future outcome is not "AI banned" or "AI everywhere." It is a more regulated split:

• AI allowed for certain workflow or language-support tasks
• AI prohibited for confidential manuscript upload into public tools
• more pressure for disclosure, auditability, and safe in-house systems

For years, peer review relied on a quiet fiction: that reviewers somehow become good at reviewing simply by being researchers.

The evidence has never really supported that assumption.

A 2025 Scientific Reports survey found that many participants had never been invited to formal training and many had never received editor feedback on the quality of their review. The same study found strong associations between reviewer self-assessment and the perceived importance of editor feedback, author feedback, and formal training.

At the same time, journals are starting to formalize what used to happen informally or invisibly.

Nature in 2025 launched a co-reviewing project in which invited referees can bring in an early-career researcher to prepare a joint report. Nature Methods announced a formal co-reviewing initiative to recognize early-career reviewer contributions. Nature Structural & Molecular Biology also rolled out formal co-review participation.

This matters for two reasons:

1. it expands the reviewer pipeline
2. it turns ghost reviewing into something more accountable and developmental

That is good for the system if editors manage it well.

The old model of peer review assumed there would always be enough qualified people willing to review out of obligation or habit.

2026 publishing no longer takes that for granted.

Publishers are trying to reduce duplicated effort through:

• manuscript transfer networks
• reviewer-report portability
• preprint-linked review visibility
• better reviewer matching systems
• internal AI tools for screening and reviewer identification

Nature Communications, for example, allows authors to transfer referee reports from another Nature journal and have those reports considered by the editors. Elsevier and Wiley both frame transfer as a way to avoid wasting reviewer labor on repeat reviews of the same paper in slightly different venues.

This is not glamorous, but it is one of the clearest signs of where the system is going. The future of peer review is not only about better reviews. It is also about fewer duplicated reviews.

A fair assessment should acknowledge genuine improvements.

More transparency

Readers can increasingly see reviewer reports and rebuttals at major journals.

More formal recognition

Named reviewers, co-reviewer recognition, and published acknowledgements are becoming more common.

More explicit AI policies

The rules are still evolving, but at least major publishers now have them.

More serious integrity investment

The system has finally admitted that fraud prevention is not a side task.

More flexible transfer and portability

Some manuscripts lose less momentum after rejection than they would have five years ago.

The improvements are real, but the friction points are still obvious.

Reviewer training remains weak

The system still depends heavily on volunteer labor with uneven preparation.

Policy is moving slower than behavior

Researchers are already using AI more widely than many journal policies seem designed to handle.

Transparency remains partial

Even where peer-review files are published, editorial deliberation usually remains hidden.

Workload is still lopsided

Highly active and visible researchers still bear disproportionate review demand in many fields.

Authors still experience opacity

The process may be more instrumented internally, but a lot of portals still reduce complex editorial workflows to vague labels like "under review."

Authors do not control peer-review policy, but they should react to where the system is going.

Write as if integrity checks are stronger

Because they are. Clean figures, explicit methods, and consistent disclosures matter more now.

Assume AI use will be judged

If you used AI in writing or analysis in ways that require disclosure, disclose it cleanly. Do not improvise.

Treat the response letter as a semi-public document

At journals using transparent peer review, it may effectively become one.

Prepare for portability

If a paper is rejected, ask whether reviewer reports can help at the next journal instead of assuming the process resets to zero.

Respect the editor's role

The modern system is more editor-shaped than many authors think, especially under stronger integrity and transparency expectations.

If you need help before or after review, compare this page with The Complete Guide to the Peer Review Process, How to Respond to Reviewer Comments, and Manusights AI Review.

Peer review is no longer just anonymous expert judgment. It is now a layered system that mixes:

• human scientific assessment
• integrity screening
• policy enforcement
• transfer logistics
• transparency choices
• and increasingly, AI governance

That makes the process messier in some ways, but also more honest about what it really is.