How AI Listens to Neurodivergence - A Comparative Study of Advice Responses Generated by Four Large Language Models

Large language models (LLMs) are increasingly being used to provide advice on a wide range of areas in life, yet little is known about how their advice changes when users supply neurodivergence-related context. Drawing on the author’s positionality as a student with autistic experience, this study examines a fixed body of 180 responses generated with ChatGPT, Gemini, Claude, and DeepSeek. to 15 adolescent-oriented advice questions under three conditions: 1) no additional context, 2) a broad neurodivergent label, and 3) specific executive-function and sensory-overload difficulties. Archived lexical indicators and close reading show that the broad neurodivergent label primarily changed tone and explanation, while framing with specific difficulties produced more concrete support strategies, with increased scripts and more concrete support strategies for task scaffolds, sensory adjustments, routines and accommodations. Variations between AI systems also appeared; however, responses spanning the data body were not generated under identical conditions or under a singular protocol, so these contrasts are descriptive and observational rather than controlled estimates of model performance. Overall, the findings suggest that specific prompts make access needs more visible than identity labels alone, potentially helping neurodivergent users obtain more targeted advice from generative AI.