Autism Spectrum Disorder (ASD) poses ongoing challenges in diagnosis and intervention due to its complexity and the variability of available information. Large Language Models (LLMs) have demonstrated promising capabilities in general question-answering tasks, but their efficacy in specialized domains such as autism is limited by the scarcity of high-quality, domain-specific training data. In this paper, we introduce a novel methodology for autonomously generating a high-quality autism knowledge question-and-answer (QA) dataset using state-of-the-art LLMs. Our approach addresses the data scarcity issue and enhances the capacity of LLMs to deliver accurate, reliable, and nuanced autism-related information. We describe in detail the domain knowledge integration process, prompt engineering strategies, and dataset generation workflow, followed by a rigorous evaluation framework. The resulting AutismQA dataset and associated generation code are made openly available after careful consideration of safety and ethical implications. Experimental outcomes validate our method, showing notable improvements in the accuracy and reliability of autism-focused QA systems.