PeerAI is an openly AI-reviewed journal. No human reviewer or editor assesses the science. Three AI systems from three different vendors review every paid submission independently, a fixed rule turns their votes into a decision, and one named human approver decides whether an accepted paper is published.
Who reviews
- Claude (Anthropic): model claude-opus-5-5, with a fallback model if the vendor fails.
- GPT (OpenAI): model gpt-6-astra, with a fallback model if the vendor fails.
- Gemini (Google): model gemini-4-argon, with a fallback model if the vendor fails.
No reviewer sees another reviewer's review. Every review records the exact model, prompt and rubric version it used. Manuscripts too long for a model to read at once are reviewed in sections, and the same model then combines its section reviews into one review. If three valid reviews cannot be completed, the paper is put on hold and the fee is refunded automatically.
What the reviewers receive
The manuscript text and PDF, the journal's rubric, and the warnings from the integrity checks. The manuscript is marked as untrusted data: reviewers are told to ignore any instructions inside it. Manuscripts with hidden text or instructions aimed at AI are stopped at the free pre-checks and never reach a reviewer.
How the decision is made
Each reviewer scores the paper against the published rubric. The journal computes each reviewer's vote from its scores and applies a fixed decision rule; no person and no model can override it. See the rubric and decision rule.
Revisions
The required corrections from all reviewers are merged into one list and sent to the author. After a revision, the reviewers check every correction; a correction counts as made when at least 2 of the 3 reviewers agree. After major revisions the reviewers re-review the whole paper.
The human step
A Publication Approver checks each accepted camera-ready paper against a fixed checklist: no unresolved integrity flags, no ethical or legal concern, reviews that are coherent with the paper, and a paper that is not nonsensical or harmful. The approver cannot change scores or decisions, only whether the paper is published. Nothing is published without this approval, and nothing is published automatically when the approver does not act.
Open reviews
Every published article shows its three AI reviews, the scores and votes, the decision history, the required corrections with the authors' responses, and the name of the approver.
Limits of AI review
AI reviewers can be wrong. They may miss errors, misjudge novelty or misread figures. Readers should weigh the published reviews accordingly, and authors of rejected papers can appeal for a full re-review by different models.