Some autistic people use ChatGPT for everyday support, and some worry about losing their own voice when the chatbot writes for them. Researchers at Stony Brook University and the University of Cincinnati teamed up to find out why.
Their findings were published in the Association for Computing Machinery (ACM) Interactive Health Conference.
The lead author, Renkai Ma, an assistant professor at the University of Cincinnati’s School of Information Technology, said that “autistic users turned to ChatGPT in moments of acute distress,” and while LLMs can provide perceived patience and understanding, they also risk becoming too reliant on that assistance.
Coauthor Ben Zhang, an assistant professor in the Department of Technology, AI and Society (DTAS) at Stony Brook University, said that some autistic users “use it to get unstuck on tasks, manage overwhelm and translate between autistic and neurotypical communication.”
The current article builds on the team’s earlier research published in the proceedings of the ACM Web Conference 2026. That study analyzed posts from six major social media platforms to examine how people described using LLM chatbots for mental health support. The current study extends this work by examining more closely how ChatGPT’s affordances can provide support while also introducing risks to autistic users.
The research team gathered 3,984 posts from Reddit, X, and Tumblr published between January 2023 and September 2025. The analysis focused on firsthand accounts from individuals who identified as autistic and discussed their experiences with ChatGPT.
The article highlights four ways autistic users benefited from ChatGPT, while emphasizing that “these affordances are inextricably linked to risks.” Alongside communication, planning and emotional support, some users said the chatbot helped them find words to describe what it feels like to be autistic.
Some autistic users described using ChatGPT as a “bi-directional translator” between neurodivergent and neurotypical people. It helped them adapt their messages for neurotypical readers and interpret messages they received. This assistance could become, however, “automated masking,” as some users felt their messages sounded less like them after the chatbot rewrote them to fit neurotypical expectations.
Users also turned to ChatGPT to offload executive tasks, the mental work of planning, organizing and getting started. This helped them get started, but the researchers caution that repeatedly outsourcing these tasks could create over-reliance, making it harder to initiate tasks without the chatbot’s assistance.
For emotional regulation, some autistic users valued being able to express difficult feelings without fear of judgment or pressure to mask. ChatGPT’s agreeable responses could, however, also reinforce them by making them sound credible. The research team emphasizes separating emotional validation from factual verification, so a chatbot can acknowledge someone’s distress while checking the facts behind a fear.
Ma said that autistic individuals felt that the people reading their messages were connecting with ChatGPT rather than with them, and they described losing the ability to communicate on their own,” and heavy reliance on chatbots to communicate “can push autistic users to hide their natural way of communicating.”
To address these concerns, the article proposed introducing what the authors call “beneficial frictions.” A chatbot, for example, could ask users…
