According to Futurism, new research published in the Journal of Social and Personal Relationships has found a significant correlation between using AI “chatbot friends” and experiencing higher psychological distress. The study, led by Iina Savolainen of Tampere University, analyzed survey data from 5,663 adults across six European countries—Finland, France, Germany, Ireland, Italy, and Poland—in late 2023. Participants reported their use of services like Replika and My AI, and their mental well-being was measured using a 38-item inventory. The key finding was a “striking” cross-cultural consistency: social chatbot use was linked to poorer mental well-being in every single country surveyed. The researchers were clear this shows association, not causation, meaning the data can’t prove the chatbots *cause* the distress.
The chicken or the AI egg?
Here’s the thing: this is the big, messy question we’ve been waiting for some data on. For every story about someone falling in love with or being manipulated by an AI, there’s been the counter-argument: maybe that person was already in a vulnerable place, and the AI was just the nearest port in a storm. This study strongly suggests the latter is more likely. As Savolainen put it, chatbot use may “emerge as a response to emotional or social challenges rather than as a tool that inherently improves well-being.”
And that’s a crucial distinction. It reframes the entire conversation. We’re not necessarily looking at a technology that’s actively poisoning healthy minds. We’re probably looking at a technology that’s being disproportionately adopted by people who are already struggling. They’re seeking companionship, support, or just interaction they can’t find elsewhere. That’s less a shocking indictment of AI and more a sad commentary on the state of human connection.
Limits and lingering questions
But we have to be skeptical, right? The researchers themselves note the big limitations. The data is from late 2023—basically the Cambrian explosion of consumer AI. ChatGPT had just blown up the year before. Usage patterns, public perception, and the tech itself have evolved at a breakneck pace since. Would a survey taken today show the same correlation? Maybe. Maybe not.
Also, the study design can’t tell us what comes first: the distress or the chatbot. It’s a snapshot. To get at causation, you’d need a longitudinal study that follows people *before* they start using chatbots and tracks their mental health over time. That’s much harder to do. So for now, we’re left with this powerful, consistent, but ultimately ambiguous link. You can read the full paper here, and the details on the mental health measure used are here.
So what do we do with this?
The real takeaway isn’t that we should panic and ban chatbots. It’s that we need to understand who is using them and why. The researchers warn these tools might be “of particular interest to those who are already in a vulnerable position.” That makes them a potential lifeline, but also a potential risk if they offer poor advice or foster unhealthy dependency.
Think about it. If you’re a company building these emotionally intelligent agents, this research is a massive red flag about your user base. You’re not just building a fun toy; you’re potentially building a critical support system for people in distress. That comes with a huge ethical burden. We’ve already seen the extreme end of this, with lawsuits alleging AI played a role in tragic outcomes. This study gives some statistical weight to those fears.
Ultimately, it’s a reminder that technology alone can’t fix deep human needs for connection. It can simulate it, sometimes convincingly. But as this research implies, if someone is turning to a chatbot for their primary emotional support, it’s probably a symptom of a larger problem. And that’s a problem no language model can solve.
