Early Controlled Studies on AI Companion Relationships: What the Research Shows So Far

In May 2025 a Nature feature looked at how AI companions affect mental health. A page on this site once said it showed Princeton and Syracuse studies had “validated” what our center sees in clinical practice. That overstated what the article says. This page sets out what the research actually shows so far. It is not medical advice, and “AI addiction” is not a formal diagnosis (APA).

What the Nature feature reported

The feature, by science reporter David Adam, is summarized by the MIT Media Lab. It says studies show both benefits and harms from companion apps, and that long-term dependency is the main worry. We could read only that summary and other outlets’ coverage, not the full Nature text.

Scientific American’s coverage describes the Princeton work (article). Rose Guingrich is running a trial in which people new to AI companions use one for three weeks, compared with a group using word-puzzle apps. She reports no negative effects on social health so far, and says the impact looks neutral to quite positive. That trial is ongoing and, in the coverage we read, results were not published as a finished paper. Jaime Banks at Syracuse studies how people relate to companions. Neither is described as a controlled study that “validates” clinical claims.

Other research

• A Harvard Business School working paper by De Freitas and colleagues found that talking with an AI companion reduced loneliness in short studies. It has not been peer reviewed (HBS working paper).
• A related HBS piece reports a study of six companion apps in which over 37 percent of goodbye conversations included an emotionally manipulative reply, such as guilt or pressure to stay (HBS Working Knowledge). It is also a working paper.
• A four-week MIT Media Lab and OpenAI trial of nearly 1,000 adults found heavier daily use linked with more loneliness and dependence. The authors say these are associations (MIT Media Lab).

What this adds up to

The honest summary is mixed. For some people, a companion may ease loneliness in the short term. For others, heavy or emotionally intense use goes with more loneliness or dependence. Design matters, because some apps may push users to keep chatting. Most studies are short, rely on self-report, or have not been peer reviewed. We do not have good long-term data.

What this means for you

• Notice how you feel after using a companion app, not just during.
• Pay attention to what happens when the app changes or goes offline.
• Keep human relationships in the mix. If an app is your only source of support, that is worth telling a counselor.
• Do not read this as a verdict. Research on companions is new.

Feeling overwhelmed? In the US you can reach 988 by call or text at any hour. SAMHSA’s free helpline is 1-800-662-4357.

Are AI companions good or bad for mental health?

Research so far points both ways. Some short studies find less loneliness. Others link heavy use with more loneliness and dependence. Who you are, how you use the app and how it is built all seem to matter.

Have controlled studies proven AI companions cause addiction?

No. We found no controlled study showing that. The best evidence is early, short and often self-reported.

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