Principal Investigator
Dora Demszky – Stanford University and Julie Cohen – University of Virginia
Project Description
This study develops and evaluates AI tools designed to support teachers in providing high-quality, personalized writing feedback aligned with their pedagogical goals. The research includes three phases: first, building a large-scale dataset of teacher feedback to characterize teacher feedback practices at scale; second, developing and experimentally testing a light-touch AI tool that provides real-time suggestions to improve feedback quality; and third, training and evaluating large language models that can generate feedback aligned with individual teachers’ styles and instructional goals. The study uses a combination of teacher surveys, experimental data from a randomized controlled trial, expert ratings of feedback quality, and student evaluations to assess the effectiveness of AI-supported feedback. The project aims to improve the quality and consistency of teachers’ feedback on student writing, while reducing the time burden on teachers and identifying design features that support effective implementation.
Research Questions
- What are the key characteristics of high-quality teacher feedback on student writing, and how do these vary across teachers and contexts?
- To what extent do AI-supported tools improve the quality of teacher-provided feedback on student writing?
- Can AI models learn to generate feedback that aligns with teachers’ instructional goals and styles, and how do teachers evaluate and adapt outputs from these models?
