As AI writing tools evolve from fixing surface errors to creating language with writers, new capabilities raise concerns about negative impacts on student writers, such as replacing their voices and undermining critical thinking skills. To address these challenges, we look at a parallel transition in university writing centers from focusing on fixing errors to preserving student voices. We develop design guidelines informed by writing center literature and interviews with 10 writing tutors. We illustrate these guidelines in a prototype AI tool, Writor. Writor helps writers revise text by setting goals, providing balanced feedback, and engaging in conversations without generating text verbatim. We conducted an expert review with 30 writing instructors, tutors, and AI researchers on Writor to assess the pedagogical soundness, alignment with writing center pedagogy, and integration contexts. We distill our findings into design implications for future AI writing feedback systems, including designing for trust among AI-skeptical educators.

Figure 1: Text presentation style is a key design element of AI writing tools.This paper explores the possible impacts of five text presentation speeds (i.e.streaming speeds) on the perceptions of tools and outputs.See Table 1 for a more precise example of each speed.Figure text adapted from Alice's Adventures in Wonderland (Lewis Carroll).Clip art figures were generated using DALL•E.