AI emotional support is better only when chosen, but shifts preferences even when it is not
While AI emotional support is rated superior only when users actively choose it, repeated interactions with AI—particularly in personal contexts—shift preferences toward AI and away from human connection regardless of initial choice.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
When people feel overwhelmed, sad, or anxious, they face a quiet but increasingly common dilemma: who do they turn to for comfort? For generations, the answer was almost always another person—a friend, a family member, or a therapist. Today, a new option has appeared on the shelf: artificial intelligence. Modern chatbots can listen without interrupting, offer words of encouragement, and never get tired. But a critical question remains unanswered by simple observation: does the quality of that comfort depend on whether the person actually wanted to talk to a machine, or are they simply stuck with it because no human was available? This question sits at the intersection of psychology and technology, exploring how our expectations shape our reality. If we believe a machine cannot understand us, does it fail us? And if we accidentally talk to a machine when we wanted a person, does that experience change our mind about who we should trust next time?
A team of researchers set out to untangle these threads through a series of four studies involving nearly two thousand people. They began by asking participants to recall a recent emotional event, something significant that had happened in the last month. Before any conversation took place, the researchers asked these people to imagine sharing that story with two different listeners: a human stranger and an artificial intelligence. The results revealed a deep divide in how people view these two options. Those who preferred to talk to a human believed that people were better at providing emotional support and that humans and machines were equally likely to judge them. In contrast, those who preferred the AI believed that humans and machines were equally good at offering support, but that the machine was far less likely to judge them. This initial belief system acted as a filter, shaping how people expected the interaction to go.
The researchers then moved from imagination to reality. In two controlled experiments, participants were asked to choose who they wanted to talk to, but then they were randomly assigned to speak with either a human or an AI, regardless of their choice. This created four distinct groups: people who chose a human and got a human; people who chose an AI and got an AI; people who chose a human but got an AI; and people who chose an AI but got a human. The findings were striking. When people received the support source they had chosen, they rated the experience highly. However, the magic of the AI was conditional. The artificial intelligence was rated as superior to the human only when the participant had explicitly chosen to talk to the machine. When people who wanted a human were forced to talk to an AI, they did not rate the machine as better than a human; they simply rated the experience lower than if they had gotten what they wanted. The superiority of AI in these short, empathetic exchanges was not a universal fact, but a result of the user's willingness to engage with it.
Perhaps the most profound discovery emerged after the conversations ended. The researchers asked participants who they would prefer to talk to the next time they had a big emotional problem. The answer depended heavily on who they had just spoken to, not just who they had wanted to speak to. People who had a five-minute conversation with an AI were significantly more likely to choose an AI for their next emotional crisis, even if they had originally wanted a human. Conversely, those who spoke with a human were more likely to stick with humans. This suggests a path-dependent shift, where the act of using a tool changes the user's future preferences. It is as if the experience itself rewrites the map of what is possible. The more people talk to the machine, the more they believe it is capable of understanding them, and the more likely they are to return to it.
To see if this effect held up in the real world, outside of a laboratory, the researchers collaborated with a major technology company to run a month-long study. Nearly a thousand people were asked to chat with an AI every day for five minutes. Some were told to talk about personal, emotional topics; others were told to discuss non-personal matters; and a third group could talk about anything they wanted. Over the course of twenty-eight days, the researchers tracked how their preferences changed. The data showed a clear trend: daily interaction with the AI, especially when the conversations turned personal, gradually shifted people's preferences. By the end of the month, more people said they would choose an AI to discuss personal issues, and fewer said they would choose a human. This shift happened even for those who started the study preferring human connection. The simple act of daily engagement with the machine updated their beliefs about its capabilities and pulled them toward it.
The study also looked closely at what made these conversations effective. It turned out that the specific behaviors during the chat mattered immensely. When the AI validated the user's feelings, expressed affection, or showed empathy, the user was much more likely to choose the AI again. Similarly, when users shared their own problems and feelings, they became more attached to the machine. The researchers found that the content of the conversation, particularly the emotional depth, was the engine driving this change in preference. It was not just the novelty of talking to a robot that changed minds, but the genuine feeling of being heard and supported during those daily exchanges.
These findings paint a complex picture of the future of human connection. They suggest that the rise of AI emotional support is not just a matter of technology becoming better at mimicking humans, but a psychological process where the experience of using the technology changes the user's mind. The preference for AI is not static; it grows with use. If a person stumbles into a conversation with an AI because no human is available, that single interaction can nudge them toward choosing the machine again in the future. Over time, this cycle could reshape how people seek comfort, potentially leading them away from human connection and toward digital companions. The researchers note that while this shift is measurable, it is not yet a foregone conclusion that people will abandon humans entirely. However, the data confirms that the path is open, and the direction of travel is influenced by the very act of taking the first step.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.