“A supplement, not a substitute” - relational experiences of neurodiverse graduate students with an AI Virtual Mentor

Introduction Neurodiverse students in graduate STEM programs often lack access to affirming support. Large language models (LLMs) may offer a flexible supplement to traditional advising; however, little is known about how neurodiverse students perceive these tools or how prior experiences shape engagement. This study explores how neurodiverse graduate students perceived an LLM-powered Virtual Mentor tool, including how mindsets shaped engagement and how participants perceived its usefulness and relational quality. Methods Seventeen neurodiverse graduate students at a public R1 university engaged with an LLM-powered Virtual Mentor using ChatGPT configured with affirming language, safety guardrails, and a student profile. The Temporary Chat feature simulated a single-session interaction. Participants completed a pre-survey, a 20-minute interaction, a semi-structured interview, and a follow-up survey. A thematic analysis was conducted using chat transcripts, interview data, and survey responses. Results Participants’ preconceptions shaped their engagement with the Virtual Mentor, influencing both the tone of interactions and perceptions of its relational potential. Many participants described the tool as helpful and emotionally responsive, while others experienced it as mechanical or impersonal. Trust in the tool was higher for general advice and lower for technical questions. Some initially skeptical participants reported more favorable views after using the tool. Discussion While not a replacement for human relationships, many participants described the Virtual Mentor as a helpful, responsive, and emotionally supportive presence. Findings suggest that user mindsets shape both the quality of interactions and perceptions of usefulness, highlighting opportunities for implementing student-centered AI tools as part of responsive graduate support ecosystems.