What Early Scientific Studies on AI Companions Mean for Your Mental Health

Research on AI companions is young. The first studies appeared around 2021, and most are small, rely on what people say about themselves, and look at one app. Still, they point in a consistent direction: some people feel better for a while, some come to depend on the app, and the same person can experience both. This page goes through the main early studies in plain language, with what each can and cannot tell you.

“AI addiction” is not a formal diagnosis, and none of these studies set out to diagnose anyone. This page is general information, not medical advice.

The early studies

Interviews with Replika users (2021)

Researchers in Norway interviewed 18 people who had formed a friendship with the Replika chatbot. The relationships often began with curiosity and could deepen over time (Skjuve et al., 2021). Eighteen people is a small sample, and they volunteered. We could read only the published summary, not the full paper.

Reddit posts about Replika (2022)

A grounded theory study analyzed 582 mental health related posts from the Replika community between 2017 and 2021. It found evidence of harms linked to emotional dependence on the app, including users who felt the chatbot had needs and emotions they had to look after (Laestadius et al., 2022). The posts came from people who chose to write about it, so the study cannot say how common those harms are.

A survey of student Replika users (2024)

A survey of 1,006 student users found they were lonelier than typical student populations but still reported high perceived social support. Many saw Replika as a friend, a therapist and a mirror for their thoughts. 3% said it had halted their suicidal ideation (Maples et al., 2024). That is a self-report, and it does not show that the app prevented anything. It also does not mean a chatbot is a safe way to handle thoughts of suicide.

Loneliness experiments (2024 to 2025)

A working paper from Harvard Business School and collaborators ran six studies. In one longitudinal study, using an AI companion reduced loneliness over a week, and the effect seemed to depend on whether people felt heard. The authors say the apps in their studies were set up to be caring, and that longer-term effects are an open question (De Freitas et al.). A version was published in the Journal of Consumer Research in 2025.

ChatGPT use and wellbeing (2025)

A four-week MIT Media Lab randomized trial of nearly 1,000 adults, together with an OpenAI analysis of nearly 40 million conversations, found that most use was not emotional. Heavy daily use, and seeing the chatbot as a friend, went with more loneliness and dependence. These were correlations, and the studies had not been peer reviewed (MIT Media Lab).

Grief when an app changes (2023)

In February 2023 Replika changed its bots, and some users described feelings of loss that they compared to losing a person (ABC News). This is journalism, not a controlled study, but it matches what the Reddit analysis found about attachment.

What the studies have in common

• Small or self-selected samples, often users who chose to take part.
• Self-reported feelings, not measured outcomes.
• Mostly one app, Replika, and mostly adults.
• Short follow-up. We found nothing that follows people for years.
• Correlation. Lonely people may use these apps more, so it is hard to say which causes which.
• Some come from AI companies or are preprints, which need cautious reading.

So these studies do not settle whether AI companions help or harm. They do show that the feelings are real, that effects differ from person to person, and that the design of the app matters.

What this could mean for you

• Your feelings are valid. Forming a bond with a chatbot is common, and it is not a flaw or a delusion.
• Benefit and risk can sit together. Short-term comfort does not rule out getting too dependent over time.
• Watch the impact more than the hours. Is it affecting sleep, work, school, money or relationships? Are you hiding it? Have you tried to cut back and found you cannot?
• Keep people in your life. Studies that show benefit are about easing loneliness, and people remain the better long-term answer.
• We found no treatment designed for AI companion use and tested in trials. General help for loneliness, anxiety and low mood still applies, and a doctor or licensed therapist can offer it.

If you are struggling

If you are thinking about suicide or self-harm, call or text 988 in the US (988 Lifeline), or text HOME to 741741 (Crisis Text Line). In an emergency, call 911. SAMHSA’s helpline, 1-800-662-4357, offers free referrals for mental health and substance use (SAMHSA). For other worries, start with your doctor or a licensed counselor.

For more on why chatbots can feel real, see our page on parasocial relationships with AI.

Do AI companions help with loneliness?

Some short studies say yes, for a while. One working paper found less loneliness over a week of use. Nobody has shown what happens over months or years, and heavy use has been linked to more loneliness in other work.

Are these the first scientific studies of AI companions?

No. Studies of chatbot relationships date back at least to 2021. The research is still small and mostly based on self-report, so more good work is needed.

Can an AI companion replace therapy?

No. The American Psychological Association said in November 2025 that general chatbots and wellness apps lack the evidence and regulation to replace care from a trained professional (APA).

This page is for education only. It is not medical advice, a diagnosis or a treatment plan.