Reviewer3 is an AI peer review tool from Reviewer3, Inc. that runs objective, verifiable checks on research manuscripts. It breaks a manuscript into its claims and checks each against the evidence offered for it, while a separate chain extracts every citation and looks for a real source. The checks cover references that are hallucinated, retracted, irrelevant or self-cited, fatal flaws such as unsupported conclusions and impossible numbers, claim review, and code replication, which runs the code and compares its output with the paper. A submission screening check, for coursework, partial drafts, leftover AI prompts and rewrites of published work, is available to editors. Authors use Reviewer3 for pre-submission review, reviewers use it for a second look and help drafting a report, and editors use it for first-pass triage.
Benefits & outcomes
- Reviewer3 reports 98.5% accuracy in separating real citations from AI-fabricated ones, tested on a dataset of real and fabricated citations with known answers.
- Reviewer3 reports that 90% of its findings “could undermine a major claim”.
- Reviewer3 reports that it ranks first among AI reviewers in agreement with humans.
Publisher / company: Reviewer3
Date of case study: Not stated
Status: Live
Built by: Company
Sources: