NEJM AI's invitation-only Fast Track review process puts AI-generated peer review alongside human review, with an initial decision promised within 7 days. A human editor first writes a complete review independent of AI; the editor then uses two LLMs (secure GPT-5 with Thinking and Gemini 2.5 Pro) to generate structured reviews; a statistical editor works with the LLMs via a structured six-step conversational process. The first two articles published this way (both RCTs of ambient AI scribes) appeared alongside an editorial describing the process, with the human and AI reviews published as supplementary material.
Benefits & outcomes
Human and AI reviews contained "both overlapping and complementary critiques"; published articles were revised substantively based on both.
AI highly effective at checking adherence to reporting guidelines and alignment between reported analyses and the Statistical Analysis Plan — time-consuming but critical checks.
Invitations limited to articles multiple editors judge likely to be accepted; accept/reject decisions remain human.
Notes & quotes
Key lesson: unstructured LLM engagement "generated multiple unsolicited critiques and suggestions that […] were sometimes off topic" — the six-step process was introduced "to establish GPT-5 as an assistant rather than a reviewer."
Clarke & Esposito's analysis (alongside the PLP pilot): AI helps with structured tasks; keep it away from interpretive judgment — https://www.ce-strategy.com/the-brief/humain-peer-review/
Publisher / company: NEJM Group
Date of mention: 18/12/2025
Source: https://ai.nejm.org/doi/full/10.1056/AIe2501175