Scientists increasingly share their results as preprints — papers posted online before peer review — and critics worry that unvetted findings may not survive closer scrutiny. A sweeping new analysis offers a reassuring answer: for the vast majority of biomedical studies, the headline conclusion barely changes on the way from preprint to published paper.
Ruslan Rust, a neuroscientist at the University of Southern California, and a colleague assembled every bioRxiv preprint posted between 2018 and 2025 that they could match to its eventual peer-reviewed version — 72,644 pairs in all. They then used a large language model to pull the main claim from each abstract and judge how much it shifted after peer review. On a checked sample, the model's verdicts agreed with two human experts about as often as the experts agreed with one another.
Small edits, not U-turns
The central claim stayed the same in 39.9 percent of cases and saw only minor revision in another 50 percent; just over 10 percent were substantially rewritten. Where wording did move, it tended toward caution: claims became more tentative after review twice as often as they became more confident. Major changes were commoner when peer review dragged on, and grew rarer over the years — from 17 percent of 2019 papers to under 6 percent of those posted in 2024.
The study, itself posted to bioRxiv and not yet peer reviewed, also looked at retractions. Biomedical papers that never appeared as preprints were pulled at roughly twice the rate of those that did — though the authors stress this is an observational comparison based on relatively few retractions, not proof that preprinting prevents mistakes.
Not everyone is fully convinced. Some researchers note that preprints carry a selection bias, shaped by who chooses to post and what they post, and one scientist cautioned that the falling rate of major revisions might reflect overstretched reviewers rather than sturdier science. Still, the message is encouraging for a research culture that now moves at preprint speed: going online before formal review, the analysis suggests, rarely means getting the science badly wrong. As the authors put it, the move to publication leaves the central claims of most biomedical abstracts intact.