An AI preprint quality score can flag strong studies before journals do. QED, a review assistant from Oded Rechavi’s team, rates originality and validity in preprints while hiding author and institution. In tests with thousands of papers, expert judges agreed with the scores about 75 percent of the time. Scientists are drowning in new papers. Peer review can take more than a year, and many teams need early signals to decide what to read or fund. QED offers one such signal. It reads a preprint, checks the claims, and gives a simple score for novelty and soundness. The goal is to highlight solid work faster and with less bias.

Why an AI preprint quality score matters

The bottleneck in science

Peer review is slow. Some biomedical papers take up to 18 months to publish. Reviewers spend late nights checking data. With so much output, most people default to journal rank as a shortcut, even though it can reflect prestige more than quality.

Reducing noise and bias

QED was built to help at the preprint stage, when journal labels and citation counts do not exist. It reviews text and figures, then scores validity and originality. It hides author names, institutions, and countries, which can lower location and gender bias that creep into human review.

How the AI preprint quality score works (QED)

Blind to prestige

The system ignores who wrote the paper and where they work. It looks only at what the manuscript claims and how well the evidence supports those claims.

What it measures

– Validity: Are the methods and controls appropriate? – Originality: Is the idea new or a small step? – Gaps: What is missing or unclear?

Scale and findings

– QED scored 57,455 bioRxiv preprints from May 2025 to April 2026. – It highlighted the top one percent across life science fields. – Country patterns were mixed: the US posted the most preprints, while Austria stood out for a high share of top-scoring work.

Does it match expert judgment?

Head-to-head with journals and reviewers

Rechavi’s team compared QED Scores for about 5,000 preprints, nearly 3,000 of which later appeared in journals. Scores showed a positive link with the eventual journal rank, and in pairwise tests where score and journal status disagreed, independent experts chose the higher-scored paper about 75 percent of the time. In other words, when journals and the score sent mixed signals, experts tended to side with the score.

Where it can miss

Sometimes a low-scored paper lands in a high-ranking journal, or the reverse. Reviewers also noted that QED can suggest extra experiments that are not realistic on tight timelines. These cases show why human judgment still matters.

Benefits, limits, and good use

What it helps with

– Fast triage of crowded reading lists – Early feedback for authors before submission – More level playing field by hiding prestige signals – A common, transparent starting point for discussion

What to watch for

– One number cannot capture all value in science – Do not use a single score to judge careers or funding alone – Keep disclosure rules for AI use in peer review – Apply practical judgment to any suggested experiments

Putting QED to work in your lab

Simple workflows

– Scan your weekly preprint feed. Sort by score, then read the top few in full. – Before submission, run your preprint to spot weak claims or missing controls. – In journal clubs, compare the score with your team’s ratings and discuss gaps. – For grant panels, use scores as a first pass, not as a final decision.

What this could mean for global science

Speed, access, and fairness

A fast, blinded review layer can help more people find strong work early, not just those close to elite labs. It may also surface high-quality studies from smaller countries or lesser-known institutions. But it should sit next to, not on top of, careful human reading.

The bottom line on early signals

An AI preprint quality score is a practical proxy when time is short and signal is scarce. QED shows that AI can align with experts much of the time and can curb bias by ignoring prestige cues. Use it to guide attention and improve drafts, then let human judgment finish the job.

(Source: https://www.the-scientist.com/can-ai-tools-spot-great-science-before-reviewers-do-74677)

For more news: Click Here

FAQ