Key points
- Users reported using the tool for executive function support and translating communication styles.
- Some users feared the AI rewritten text sounded less like them, creating a masking effect.
Researchers from Stony Brook University and the University of Cincinnati published findings in the ACM Interactive Health Conference detailing how autistic individuals interact with ChatGPT. The study analyzed 3,984 posts from Reddit, X, and Tumblr published between January 2023 and September 2025.A new study shows autistic people use ChatGPT for help but worry it changes how they sound.The ACM publication details how LLM affordances for autistic users create a trade-off between executive support and identity preservation. The authors focused on firsthand accounts from people who identified as autistic and discussed their specific experiences using the chatbot for various forms of support.
How we got here
The current work builds on earlier research by the same team, published in the ACM Web Conference 2026 proceedings. That previous study examined posts from six major social media platforms to understand how people described using large language models for mental health support. The new article extends that scope by looking more closely at the specific affordances ChatGPT provides and the associated risks for autistic users.
What is new
The analysis highlights four ways users benefited, including communication aid, planning assistance, emotional support, and finding words to describe the autistic experience. Lead author Renkai Ma noted that users often turned to the tool during moments of acute distress. Coauthor Ben Zhang said users employed it to manage overwhelm and translate between neurodivergent and neurotypical communication styles.
Reading between the lines
Users described the chatbot as a bi-directional translator that helps adapt messages for neurotypical readers and interpret incoming messages. However, the authors emphasized that these benefits are inextricably linked to risks. Some users felt their rewritten messages sounded less like them, a phenomenon they described as automated masking. Others worried about becoming too reliant on the AI's perceived patience and understanding.
The open questions
The study presents a tension between utility and identity preservation. While the AI helps users get unstuck on tasks and offload executive work, the evidence suggests this assistance may come at the cost of authentic self-expression. The sources do not provide quantitative data on how many users experienced negative identity effects, leaving the scale of the masking risk unclear.
Why it matters
- Evaluators must consider that accessibility features in LLMs may inadvertently encourage masking of neurodivergent traits.
- The finding that users rely on AI during acute distress highlights potential safety gaps in current model behaviors.
- The risk of automated masking suggests that output personalization may not align with user identity preservation goals.
- Understanding the bi-directional translation use case reveals a specific demand for communication mediation tools.
Topics: AI ethics, accessibility, large language models, neurodiversity, research
Updated
The strongest doubt
Counterpoint from the sources
The authors note that while risks exist, the AI provides perceived patience and understanding that users find valuable during distress.
Did this article give you the evidence you needed?
Your choice stays on this device. No public tally.Questions readers ask
What platforms did the researchers analyze for this study?
The team analyzed posts from Reddit, X, and Tumblr published between January 2023 and September 2025.
Did the study find that all autistic users experienced negative effects?
No, the study highlighted benefits like task management and emotional support, but noted risks like automated masking.
Who were the lead authors of this research?
Renkai Ma from the University of Cincinnati and Ben Zhang from Stony Brook University led the research.
Connected stories
Reading receipt
- Users reported using the tool for executive function support and translating communication styles.
- Some users feared the AI rewritten text sounded less like them, creating a masking effect.
Still open: How prevalent is the experience of automated masking among the broader autistic population?
Still open: What specific model behaviors trigger the feeling of losing one's voice?
