AI reviewers and AI co-authors
Team members backed by the AI, each with a persona and a model. They screen blind, discuss and may change their votes, extract, and draft or review sections; they are never authors.
On this page
An AI co-author (also called an AI reviewer) is a separate member of the team, powered by the AI, with its own persona: expertise, screening style, instructions and model.
What they do
- Screen blind and independently, with their own votes and a verdict on each criterion.
- Discuss: in conflict threads they explain and defend their votes, read the record (and at full text the PDF), and may change their vote with reasons.
- Extract data as a second extractor.
- Draft or review manuscript sections.
They cannot log in, and they are never listed as authors (COPE, ICMJE and IEEE policies); their contributions go into the paper's AI-use statement.
Screening style
| Style | Behaviour |
|---|---|
| Inclusive | When unsure, include: fewer missed studies, more full texts to read |
| Balanced | Apply the criteria literally |
| Strict | Include only clear matches |
A style different from yours gives a genuinely independent second opinion.
How their agreement is reported
Their votes count in the review like anyone's, and their agreement is reported separately: as AI pairs (two AI reviewers) and AI and human pairs. A vote they change in a discussion still counts with its pre-discussion value in κ. See Agreement and Cohen's kappa.
Keeping a persona for other reviews
An AI co-author normally belongs to the review it was made for: it appears on that team only and is deleted with the review. Keep it for other reviews to use the same persona on several teams. A kept persona is shared: a change to it applies in every review that uses it, and it stays when a review is deleted.
Two AI reviewers on two different models (ideally from two different makers) make the most independent pair. On one model they agree on almost everything, so their agreement says little.