Autistic users describe ChatGPT’s help—and the risk of losing their own voice

ChatGPT can help some autistic people communicate and get started on difficult tasks, yet leave them feeling that their words are no longer their own. In findings described by Stony Brook University on November 11, researchers identified benefits and risks within the same uses—not a simple divide between helpful and harmful interactions.

The team, from Stony Brook and the University of Cincinnati, gathered 3,984 posts from Reddit, X and Tumblr, published between January 2023 and September 2025. Its analysis focused on firsthand accounts from people who identified as autistic and discussed their experiences with ChatGPT. These are descriptions of use, not a clinical test of whether chatbot assistance improves health or causes dependence.

A translator that can change the speaker’s voice

Some users treated ChatGPT as a translator in both directions between autistic and neurotypical communication. They used it to adapt outgoing messages for neurotypical readers and to interpret messages they received. Others said it helped them find words for the experience of being autistic, giving language to something they struggled to describe.

But rewriting a message to meet another person’s expectations could also make it feel less authentic. Some users said the edited text sounded less like them. The researchers described this risk as “automated masking”: chatbot assistance that pushes autistic users to hide their natural communication style rather than helping others understand it.

Lead author Renkai Ma, an assistant professor at the University of Cincinnati, said some autistic people felt message recipients were connecting with ChatGPT rather than with them. He also described users reporting that they were losing the ability to communicate on their own. The concern was not simply whether a rewritten message was clear, but whose voice the recipient encountered.

Getting unstuck without handing everything over

Another use involved executive tasks—the mental work of planning, organizing and getting started. Users described offloading that work to ChatGPT to move past a block or manage overwhelm. The researchers cautioned that repeatedly outsourcing these steps could create over-reliance, making it harder to begin without the chatbot. That is a proposed risk, not a measured decline established by these accounts.

Emotional support brought a related tension. Some users valued being able to express difficult feelings without judgment or pressure to mask. Ma said autistic users turned to ChatGPT during acute distress, when its apparent patience and understanding could be especially appealing. Yet agreeable responses could also make a user’s fears sound credible, rather than merely acknowledging how upsetting those fears felt.

A pause before the chatbot makes a plan

The team’s proposed response includes what it calls “beneficial frictions.” Instead of immediately producing a plan, a chatbot could first ask users to set their priorities. That would leave the user responsible for choosing what matters before the system suggests how to proceed. The proposal addresses the planning risk without treating all assistance with getting started as a problem.

The researchers also proposed tools for neurotypical readers to interpret direct messages from autistic colleagues. That would put assistance on the receiving side of an exchange, rather than asking only the autistic sender to adapt. These ideas will guide the team’s future research; they are design proposals, not demonstrated fixes for masking or over-reliance.