New CDT Research Will Explore the Impacts of Social Relationships with AI Chatbots on Neurodivergent Young Adults

Also authored by McKynzie (Kynzie) Clark.

AI chatbots capable of and even designed for social companionship have proliferated in recent years. Ranging from generative AI platforms with chat interfaces (e.g., ChatGPT, Claude, or Perplexity) to more visually stimulating AI-powered avatars capable of text, talk, or video engagement (e.g., Replika, Character.AI, Nomi, etc), AI “companions” offer users seemingly infinite availability, friendship, mentorship, and even romance.

The once science fiction concept of an artificial friend or partner is now a reality for many, especially among young people. A recent report from Common Sense Media found that more than 70% of teens have used AI companions and about half use them regularly. The Institute for Family Studies/YouGov found that one in four adults under 40 surveyed believe that AI has the potential to replace real-life romantic relationships.

Despite this increasingly widespread use, research about long-term characteristics and impacts of social interaction with chatbots is scarce. In a recent paper, researchers identified potentially harmful AI companion traits and hypothesized several design features that could cause those traits and threaten to harm AI companion[1] users, leaving those causal connections for empirical evaluation. In a new research project at CDT, in collaboration with the Youth, Media & Wellbeing Research Lab at Wellesley College, we aim to begin to fill some of the empirical gaps by probing the longitudinal effects on young, neurodivergent adults of using AI chatbots as social companions, as well as the design choices that contribute to potential benefits and harms.

Such research is critical because we have already seen at least anecdotal evidence of harms experienced by some AI companion users. In the last year, reports of AI that allegedly encouraging delusions, self-harm, suicide, and homicide have become concerningly common.

In response, legislators have begun attempting to regulate the technology. California’s recently passed Companion Chatbots bill (SB No. 243) requires that tech companies making AI chatbots 1) warn users that the companion chatbot is not human, 2) maintain a protocol for preventing the production of suicidal ideation, suicide, or self-harm content to the user, as specified, and 3) publish details on that protocol on their website. Similarly, New York’s Artificial Intelligence Companion Models statute (effective since November, 2025) requires that tech companies develop a protocol to take “reasonable efforts for detecting and addressing suicidal ideation or expressions of self-harm” expressed by a user to the companion.

Well-intended as these proposals may be, regulating AI companions without access to a more robust body of empirical research puts the cart before the horse, leading to policy without demonstrated effectiveness. The psychological effects of interacting with companion chatbots — and the attendant effects of companies’ design choices — require further research to enable better design choices and more effective policy.

Longitudinal Research on the Impacts of Social AI Chatbots on Neurodiverse Young Adults

To address this gap, CDT will investigate the long-term impacts and effects that AI chatbots may have on young adults who use them for social support and interaction. Even less researched than the topic at large are the effects of AI companionship on people who are neurodivergent. Approximately 15-20% of the world’s population is neurodivergent, which makes this a critical group to consider when developing inclusive, evidence-based policy. More importantly, initial signals suggest that AI companionship may present neurodivergent users with different benefits and risks from neurotypical users. Further, several instances that made headlines involved neurodiverse users, although that was not the focus of the stories.

In this research we take a human-centered design approach that will focus on neurodiverse[2] young adults who have an established social relationship with an AI chatbot. Participants will complete a series of surveys and provide weekly “diary entries” that document noteworthy interactions with their AI chatbots over the course of 10 weeks, followed by an in-depth interview. By integrating survey, diary-based, and interview data, the project aims to generate insights and recommendations from neurodiverse young adults’ perspectives, that can guide the development of safe, ethical, and accessible AI chatbot and companion systems and inform policy. Some of our specific research questions include:

1. What motivates young neurodivergent people to use AI chatbots as companions? How do these interactions make them feel?
2. How does perceived loneliness/social belonging correlate to the usage of AI chatbots as companions? How does perceived strength of social relationships (human-human) correlate to the usage of AI chatbots as companions?
3. What are some of the design choices that impact neurodivergent users?
4. How can AI chatbots be more safe/accessible for neurodivergent users?

Throughout the process of designing this research project, we have sought the input of neurodivergent advocates and disability experts to ensure our study instruments are accessible. To that end, we have provided all materials in plain language, which we have defined as an eighth grade reading level or lower as calculated by the Flesch-Kincaid scale. We have also developed exemplars and step-by-step guides to accompany study instruments.

—

We need a better understanding of how young neurodivergent people use AI companions as well as how they experience benefits and risks from their use. This is a crucial inquiry to make sure that AI chatbots are safe and accommodating to all users. Through our forthcoming research, we intend to provide insights and recommendations to designers and practitioners, as well as suggest new policies that center the experiences of neurodivergent people to foster safer AI tools for everyone.

[1]  Defined here as any AI system that provides companionship for a human, regardless of its designer’s intention or whether there are also instrumental motivations for human interaction with the AI system.
[2]  We define neurodivergence as a non-medical umbrella term describing those whose brains may function, learn, and/or process information differently including some individuals with intellectual and/or developmental disabilities, such as attention deficit hyperactivity disorder (ADHD) and autism.