Peer review is the quality check scientists lean on, and for the average published preprint, the reviewed version appears about eight months after the work goes public.

In 2025, bioRxiv and medRxiv received more than 64,000 new preprints between them, and openRxiv reports that about 80% are later published in a journal, an average of 250 days after posting. 

Through those months, researchers read the work, discuss it, and plan experiments around it. They do this without the signals they usually lean on, because a new preprint has no journal name, citation record, or reviewer reports attached.

Careful readers close that gap by going to the claims themselves. They ask what each experiment shows, whether the evidence supports the conclusion, and whether the finding is new. A preprint score applies those questions to each main claim in a manuscript, with author names and affiliations removed, in minutes.

A preprint score can show where a preprint's evidence holds before any journal weighs in. Peer review still has jobs a score can't do, and both sides of that line are covered below.

What a Preprint Is and Why It Exists Alongside the Journal System

NIH describes a preprint as a complete and public draft of a scientific document, shared before peer review. Authors upload the manuscript to a preprint server, and it usually goes live within days. On bioRxiv, staff and volunteer affiliates screen each submission for scope, completeness, plagiarism, identifying participant information, and potential for public harm. That screening keeps out unsuitable material, and it makes no judgment about whether the science holds.

Researchers post for practical reasons. A preprint establishes priority, puts the work in front of colleagues months before a journal would, and invites feedback while revisions are still possible. NIH also lets applicants cite preprints wherever they cite other research products, as long as the citation gives the DOI and identifies the object as a preprint.

Funders are also moving toward open access publishing, and some now ask for preprints directly. The Gates Foundation expects funded manuscripts to be shared as preprints and no longer pays article processing charges. NIH requires that funded manuscripts accepted on or after July 1, 2025 be available in PubMed Central on the publication date, with no embargo. eLife now publishes every paper it reviews as a Reviewed Preprint, with public reviews and an assessment in place of an accept or reject decision.

Journals have largely made room for this. In journal policies collected around 2019, 91% of life and earth sciences journals allowed preprints in some form. How many preprints reach a journal depends on the sample. openRxiv puts the figure at about 80% for its own servers, while a 2026 scoping review of health-field studies, most of them limited to COVID-19 research with short follow-up, found a median publication rate of 42%.

‍

‍

What Peer Review Evaluates and Where Its Limitations Consistently Appear

Journal peer review starts with the editor, who decides whether to reject a manuscript outright or send it out for review. If it goes out, a few experts in the specific field review it in depth. They assess whether the methods are sound, whether the data support the conclusions, how significant the work is, and whether it suits the journal. The editor then weighs those reports and decides whether to accept the paper, reject it, or ask for revisions. Studies of that process point to four weaknesses:

• Reviewers often disagree: A meta-analysis of 48 studies covering 19,443 manuscripts found low agreement between reviewers, with a mean reliability coefficient of .34, a level the authors describe as quite limited.
• Major errors get through: When the BMJ planted nine major errors in a test paper, reviewers found an average of 2.58 of them at baseline.
• Author identity shapes judgment: In a randomized trial of orthopedic reviewers, those who could see that the authors were prominent rated an identical manuscript more favorably and recommended acceptance more often.
• Judgment arrives late: Formal review of a preprint happens months after the work goes public, and readers make decisions in the meantime.

None of this makes peer review optional. It marks out what a journal decision can certify, and what a reader still has to judge alone.

‍

What a Preprint Score Measures That Journal Acceptance Does Not

Journal acceptance is a single decision about a whole paper, often reached after several rounds of revision. When reviewers judge a paper’s central claims sound, they usually ask for more work to support them, such as repeat experiments, added controls, and further data analysis. 

When the claims don’t hold up, reviewers reject the paper. So the preprints that reach a journal have already passed that filter, and their central conclusions tend to survive while the evidence behind them grows. A 2026 scoping review of health-field studies found that preprints' content, outcomes, and conclusions stay largely consistent with their published versions. 

Comparisons that look past the abstract show the added work. In a study of bioRxiv and medRxiv preprints posted early in the COVID-19 pandemic, about 23% of published articles had significant content added to or removed from their figures, and none of the bioRxiv preprints lost content. An LLM-based analysis of 72,644 bioRxiv preprint–publication pairs, itself a preprint not yet peer-reviewed, found the primary claim substantially revised in 10.2% of abstracts, and its authors note that comparing abstracts misses changes to methods, figures, and results.

A conclusion can stay the same while the evidence for it changes a lot. A single accept decision doesn’t show readers which claims reviewers questioned or what evidence they asked for. Someone reading a new preprint can’t yet tell whether its claims are sound, or how much more support they’ll need.

A preprint score works at the claim level. The manuscript is anonymized first, so author names and affiliations are irrelevant. It then breaks the manuscript into its main claims and the experiments behind each one, and scores each claim for originality and validity against the existing literature. The table compares what each signal tells a reader.

‍

‍

How Preprint Scores Change the Way Researchers Evaluate Early Work

The main change is timing. A score gives readers a structured first pass on a new preprint's claims near the time it posts, alongside their own reading, without waiting months for a journal decision or relying on the authors' reputation. That helps with three everyday decisions, and in each one, the reader makes the final call.

Deciding What to Read

No one can read every new paper in their field, so the daily question is which few papers deserve a careful read. A useful filter has to weigh quality as well as topic, because a relevant paper with weak evidence still costs hours of reading time. 

QED Science's Insights Feed is an AI-powered tool that tracks new literature and preprints and surfaces the papers that fit a researcher's own profile, without the researcher having to search for them. It ranks new PubMed and bioRxiv papers down to individual claims and shows the QED Score alongside top-scoring recommendations, with a short note on why each paper matters to that reader.

Finding Strong Work Outside Top Venues

QED Score is a validated quality metric that rates a manuscript for originality and validity, built up from grades on each claim. Its validation study found that a score computed on a preprint correlated with the rank of the journal where the paper was later published (Spearman 0.63, n = 2,879), and 12.9% of matched preprints scored above the median but were published in journals ranked below the median. An anonymized score can flag work whose quality and journal rank disagree.

QED applied the same scoring in the anonymized evaluation of 57,455 bioRxiv preprints posted between May 2025 and April 2026, and selected the top 574 as The 1%.

Checking Your Own Work Before You Post

Paper Review breaks a manuscript into its core claims and shows where the evidence is thin, with specific, actionable findings, before the preprint goes public. Reports stay private in the author's account. 

QED does not use uploaded manuscripts to train its models to reproduce, rephrase, or refer to unpublished work, and nothing an author uploads can be exposed unless they choose to share it.

‍

The Limits of Preprint Scores and When Peer Review Still Applies

A preprint score is a first-pass signal with clear limits.

• It assumes the data are genuine: A score tests whether the evidence supports the claims, and it does not detect fabricated or manipulated data. Hidden text aimed at AI tools is a separate risk. In 2025, authors were found hiding prompts telling AI reviewers to give positive reviews, though one author's keyword search, posted as an arXiv preprint, found no cases on bioRxiv or medRxiv.
• Performance varies by field: QED's own validation found the correlation with journal rank was strongest in genetics and weakest in systems biology. QED's working explanation, not yet tested with experts in those fields, is that the score leans toward mechanistic reasoning, which would serve descriptive, observational, and methods papers less well.
• AI feedback supports experts: In one large analysis of AI peer review, GPT-4 feedback overlapped with the points raised by Nature-family reviewers about as often as two human reviewers overlapped with each other. The authors concluded that such feedback can help researchers while they develop a manuscript, and that expert review should remain central.
• Formal review still decides: Publication and funding decisions run through peer review. NIH bars its reviewers from using generative AI on applications, and a score belongs with authors checking their own work and readers deciding what to read.

‍

Score Your Manuscript Before You Post It

A preprint is public the moment it posts, and readers start judging it right away. A claim-level review beforehand shows which claims hold up and which need more evidence.

QED Science is free for academic researchers. Upload your own manuscript to see where the evidence holds and where it needs work, while the work is still yours to change.

‍

FAQs