A voice-and-avatar AI companion supports skill development in autistic children: a 12-week randomized controlled trial

Autism is a way of being that changes how a person experiences the world, particularly in how they communicate, process sensory information, and interact with others. For many families, the goal is not to change the child's fundamental nature, but to help them build the specific skills they need to navigate daily life with confidence and joy. Traditionally, support has come from human therapists who work one-on-one with children, but the landscape is shifting. Technology has introduced new tools, from text-based programs that adjust lessons to voice assistants that practice conversation. Yet, these tools often work in isolation, focusing on just one area of development, and many are built on the idea that autistic traits are problems to be fixed rather than differences to be accommodated. The question researchers are now asking is whether a single, integrated digital companion could support a child across many different skills at once, while respecting their unique way of experiencing the world.

A team of researchers recently tested this idea with a new kind of artificial intelligence companion designed specifically for autistic children. They created a system that combines spoken conversation with animated characters, or avatars, that appear on a tablet screen. Unlike a teacher or a doctor, the system was designed to act like a peer—a friendly guide rather than an authority figure. For younger children, the companion was an animated animal, like a dog or a cat. For older children and teenagers, it was a virtual human peer. The system listened to the child's voice, responded with speech, and used the avatar to show gestures and expressions, all while adapting to the child's pace and sensory needs. The researchers wanted to see if using this companion for three months would help children improve in five key areas: social communication, emotional awareness, thinking skills, daily living tasks, and language use, compared to children who continued with their usual care without the new technology.

To find out, the researchers worked with thirty-two children and teenagers, ranging in age from three to sixteen, all of whom had a clinical diagnosis of autism. They split the group into two equal parts. One group used the AI companion for twelve weeks, engaging with it three or four times a week for sessions lasting between fifteen and twenty-five minutes. The other group, serving as a comparison, continued with their standard therapies and support services during the same period but did not have access to the AI. Crucially, the experts who evaluated the children's progress did not know which group each child belonged to, ensuring that the results were based purely on the children's performance. The evaluation covered eighty-three specific skills, from how a child greets a friend to how they plan a daily schedule or handle money.

The results showed a clear difference between the two groups. After twelve weeks, the children who used the AI companion improved significantly more than those in the comparison group across every single skill area measured. The children using the system became much better at starting conversations, understanding how others feel, and managing daily tasks like shopping or navigating traffic. They also showed stronger gains in organizing their thoughts and using language to describe the world around them. The improvement was not just a small bump; the changes were substantial, with the AI group showing large leaps in ability compared to the group that only received standard care. For instance, in the area of daily living skills, the AI group gained an average of thirteen points on the assessment scale, while the comparison group gained less than three. Similarly, in language use, the AI group improved by more than eight points, whereas the comparison group improved by just one point.

Beyond the test scores, the way the children interacted with the system also changed in positive ways. As the weeks went by, the children became more accurate in touching the screen to select answers, their responses to the avatar's voice became faster, and they expressed a wider range of emotions during the sessions. Most notably, the number of sessions they completed successfully rose from about two-thirds at the start to nearly nine-tenths by the end. This suggests that the children were not just tolerating the technology but were genuinely engaging with it, finding the peer-like interaction compelling enough to stay focused and participate fully. The system's design, which allowed children to adjust volume and speed and did not punish repetitive movements, seemed to help them feel safe and comfortable enough to learn.

The researchers emphasize that this approach was built on a foundation of accepting autistic traits rather than trying to erase them. The system did not force children to act "normal" or suppress their natural behaviors; instead, it met them where they were and helped them build the skills they and their families wanted to learn. While the study was limited to a small group of thirty-two participants and lasted only three months, the findings suggest that combining voice and visual avatars in a peer-like format can offer powerful support for skill development. The authors note that future studies will need to involve more children and track progress over longer periods to confirm these results, but the current evidence points to a promising new direction. By treating the AI as a companion rather than a therapist, and by focusing on strengths rather than deficits, this technology offers a way to help autistic learners expand their abilities and options in a world that often demands they change who they are.