Neurodivergent Students and Generative AI: What Think-Aloud Protocols Reveal about Writing with AI

As generative AI becomes embedded in student writing, institutions face an urgent question: how do different students actually use these systems during composing? This session presents findings from a study using Think-Aloud Protocols (TAPs) with 100 undergraduate writers completing academic writing tasks while interacting with generative AI tools. The project focuses specifically on patterns among self-identified neurodivergent (ND) students, a population frequently discussed in accessibility conversations but rarely examined empirically in AI-mediated writing research. In the IRB-approved study, students completed a brief demographic survey before engaging in a recorded writing task in which screen activity and verbalized cognition were captured and analyzed by a team of faculty, graduate researchers, and undergraduate research assistants. Preliminary analysis suggests that ND students do not simply use AI to generate text. Instead, many employ AI as a cognitive scaffold: iterating prompts, externalizing planning, and reducing executive-function friction during composing. These patterns differ in notable ways from those observed among neurotypical peers, particularly in iteration frequency, task decomposition, and reflective prompting strategies. This session shares methodological insights from TAPs research and discusses how these findings can inform inclusive AI pedagogy, accessible writing support, and institutional AI policy.