XSci puts AI peer review to the test with open-review competition
XSci has launched a new preprint platform combining AI review with paid expert peer review, alongside a competition offering researchers $500 for papers that receive expert certification.
The platform is designed to add an additional trust, or certification, layer to preprints. Authors can post their research free of charge with a DOI, use named AI reviewers to assess their work, and request paid peer review from a human expert.
Reviews can be published alongside papers, at the author’s discretion, allowing readers to examine the assessments and make their own judgements about the research. Papers that successfully pass the human review process receive an ‘XSci Peer-Certified’ mark.
Authors retain all rights to their work, while the platform is free to submit to and read and is open to researchers in any discipline and country.
Alongside the launch, the company has opened the XSci Open Review Competition 2026, through which up to 100 submitted papers will receive a human expert review paid for by XSci. Authors of papers subsequently certified by the expert will receive $500 and be entered into the XSci Open Review Award 2026.
To enter, researchers must verify either an ORCID iD or an institutional email address. Submitted manuscripts are posted as preprints with a DOI and can receive reports from several named AI reviewers within minutes or hours. Authors can choose which AI reviewers to use for feedback or enter the competition without using them.
XSci said the human reviewers will assess whether a paper, following revisions of the kind normally requested by a journal, has the potential to be published in journals nominated by the authors or, where no journals are specified, leading journals in its field, provided it contains no major error.
Of the 100 human review places, 60 will be allocated on a rolling basis in submission order to entries whose AI reports identify them as candidates for leading journals in their field. The remaining 40 will be allocated after entries close, divided among disciplines in proportion to the number of submissions and then ranked within each discipline.
Researchers whose papers are not selected for the 100 funded reviews will receive $49 off XSci’s human review service for the same paper, valid until 31 March 2027.
Authors can submit papers elsewhere at any time. Papers already published in journals or conference proceedings are ineligible, although papers already available as preprints or under review elsewhere can be entered. Each author can submit up to three papers, subject to the agreement of their co-authors.
“Science has never produced so much, and never checked so little of it,” Dr Jeffrey Poon, Publishing and Editorial Director at Xsci, noted. “The bottleneck has moved. It is no longer hard to produce a paper. It is hard to know which papers to trust. We think the answer is not to slow production down but to bring the checking up to speed, and to do it in the open.”
Poon also remarked “this competition is the first public test of AI review against real papers at any scale, and that it can be seen to fail”.
XSci said it will publish details of the method used to select papers, a breakdown of entries and awards by discipline, and an assessment of where the process encountered problems and could be improved.
A method note outlining the selection signals, but not the threshold value, is due to be published on 6 November 2026. The full methodology, including thresholds and adjustments, will be released alongside the final results.
Entries close on 30 November. Results for early admissions will be released on a rolling basis, with final results expected by the end of December.