A new study asked whether generative AI can grade the graders, and the answer is: pretty well.
Researchers at a US public university analyzed 120 metareviews, AI-generated critiques of student peer reviews written in graduate online courses, using Systemic Functional Linguistics and Appraisal Theory to see how the feedback built meaning. They examined three dimensions: what the feedback said, how it related to the student, and how it was organized on the page. The AI-written reviews balanced praise with constructive criticism, stayed aligned with rubric expectations, and staged comments in a way that left room for student agency. The result reads less like a form letter and more like notes from an attentive teaching assistant.
That matters because peer review only works if students get useful feedback on their feedback, and that step usually gets skipped since instructors rarely have time to comment on every comment. If AI metareviews can reliably hit the tone and structure of solid human feedback, it fills a gap most online courses currently leave empty. It's a narrower, more useful contribution than most AI-in-grading pitches, which tend to focus on scoring essays rather than teaching students how to critique each other's work.
Worth remembering: this covers 120 metareviews at a single university, not a validated grading product, so treat "AI approximates good feedback" as a promising research finding rather than permission to let a chatbot run the gradebook.