Measured against real peer reviews, and reported either way.
Reviews of 5 manuscripts published in The BMJ between 2021-2023, scored against the peer reviews those journals published alongside them.
Against a journal reviewer average of 7.86, scored blind across ten dimensions.
Scored within one point of the published journal review for the same manuscript.
Issues raised by one reviewer and not the other, in both directions.
This is a preliminary analysis of 5 manuscripts, not a controlled trial. The complementarity figure of 64.2% cuts both ways: it means the AI panel and the journal reviewers each raised substantial issues the other missed. Read it as evidence that the two are complementary, not that either replaces the other.
The manuscripts
Every manuscript, its published journal review score, and the score the panel earned on the same paper. Open any review to read it in full.
| Manuscript | Design | Journal | Panel | Review |
|---|---|---|---|---|
| Hippisley-Cox et al. 2022 | Risk Prediction Model | 8.3 | 9.2 | Open PDF |
| Mok et al. 2023 | Pharmacoepidemiology Cohort | 7.6 | 9.6 | Open PDF |
| Morales et al. 2022 | Interrupted Time Series | 7.8 | 8.1 | Open PDF |
| Rees et al. 2021 | Population-Based Cohort | 7.8 | 8.8 | Open PDF |
| Woolf et al. 2023 | Mendelian Randomisation | 7.8 | 8.6 | Open PDF |
Full scoring rubric, dimension definitions, and the replication prompt are in the methodology.
See what it finds in yours.
Upload a manuscript and read the assessment before deciding anything.
Get free assessment