New research from Stanford University suggests AI writing tools can deliver noticeably different feedback depending on how a model is prompted to describe a student’s race, gender, and motivation. In the study, researchers fed 600 middle-school essays into four AI models and generated writing feedback. They then resubmitted each essay 12 additional times, changing only the description of the writer (for example, identifying the author as Black or white, male or female, highly motivated or unmotivated, or having a learning disability). Across models, the feedback patterns shifted in consistent ways. Essays attributed to Black students received more praise and encouragement, sometimes with emphasis on leadership or power. By contrast, essays attributed to Hispanic students or English learners were more likely to trigger grammar and “proper” English corrections. When the writer was described as white, feedback more often focused on argument structure and evidence. The study also found tone differences: AI feedback to female-labeled students was more affectionate and used more first-person pronouns, while unmotivated-labeled students received upbeat encouragement. Higher-achieving or motivated labels more frequently drew direct, critical suggestions aimed at refining work.