Preprint
A scientific manuscript posted publicly before formal peer review — enables rapid dissemination of research findings and is standard practice in physics, biology, and economics, though it creates challenges for AI tools that cannot distinguish peer-reviewed results from unreviewed claims.
What it means
A preprint is a version of a scientific paper that is made publicly available on a preprint server before (or sometimes instead of) formal peer review by a journal. Authors upload the manuscript, and it becomes accessible — sometimes within hours — with a permanent identifier (DOI) that allows citation.
Major preprint servers:
- arXiv (physics, mathematics, computer science, economics, quantitative biology) — the original and largest, founded 1991
- bioRxiv (biology, except clinical medicine) — founded 2013, now hosts millions of papers
- medRxiv (clinical medicine and health sciences) — for health research
- ChemRxiv (chemistry) — run by the American Chemical Society
- SSRN (social science, economics, law)
- ESSOAr (earth and space sciences)
Why preprints matter for AI research tools
AI literature tools have uneven preprint coverage. Semantic Scholar and OpenAlex index preprints alongside peer-reviewed papers; PubMed traditionally does not (though bioRxiv/medRxiv integration has improved). Elicit primarily draws from Semantic Scholar’s database, so it includes preprints. Tools that search only PubMed miss the preprint literature.
LLMs were trained on preprints. Large language models were trained on text from arXiv, bioRxiv, and similar servers. A model that confidently cites a “paper” may be citing a preprint that was never peer reviewed, or a preprint that was later revised significantly or even retracted. There is no reliable way to tell from an LLM’s output alone whether a cited paper has been peer reviewed.
The COVID-19 precedent. The speed of preprint dissemination during COVID-19 allowed rapid scientific communication but also led to high-profile preprints with methodological problems being widely cited before peer review could catch the issues. This tension between speed and quality remains unresolved.
Reading preprints critically
Preprints have not been peer reviewed. This does not mean they are wrong — peer review is imperfect, and many important findings appear first as preprints — but it means the reader takes on more of the evaluation burden. When reading a preprint, check:
- Is it from an established group with a track record in this area?
- Has it been cited or commented on by the community?
- Are the methods clearly described and reproducible?
- Has a journal version since been published? (Use DOI resolver or search the journal)