What AI May Teach Us About Expressing Empathy
While there are serious concerns that AI could negatively affect children’s social development and mental health, there is also another possibility worth paying attention to. For some children, AI may offer a low pressure space to practice social skills, work through difficult emotions without fear of judgment, and encounter perspectives that help them think differently about what they are going through. So, the question is not only how to protect children from harm, but also how to guide them toward uses of AI that can safely support emotional and social development.
Large language model, or LLM based AI, has shown a strong ability to generate empathetic language. Previous studies have found that in blinded evaluations, LLM responses are often judged as more empathic than human written ones. But there is an important catch. When the same kind of response is labeled as AI generated, people tend to feel less heard and less supported than when they believe it came from a human.
This suggests that even when AI can sound empathetic, people still prefer to feel heard and understood by another human rather than by a machine. That is why, if we want to emphasize human connection as a core value in society, it is critical for children and adolescents to learn how to respond to each other with empathy. The question then becomes: where can AI fit into that process?
The study
A recent study by Kumar and colleagues examined whether AI can help people improve their empathic responses when another person shares a personal or work related concern. The researchers created a platform called Lend an Ear, where people practiced offering empathic support to an AI role playing partner. The study included 968 participants, with a mean age of 45.6 years, and 2,904 text based conversations.
The results suggest that personalized AI coaching improved empathic communication across several dimensions. People became better at encouraging elaboration, validating emotions, and demonstrating understanding. They also became less likely to rely on less helpful responses such as unsolicited advice or dismissing emotions.
What stood out to me
These findings are encouraging. They suggest that personalized feedback from AI can help us express empathy to another human more effectively. But there were two parts of the findings that especially caught my attention and made me think more deeply about their implications.
When Empathy Starts to Sound More Like AI
Another type of empathic response that decreased after the AI intervention was motivational empathy, for example, You are doing the best you can or I think that is a great start. In the figure below from the study, I found it interesting that as motivational empathy, which people naturally expressed a substantial portion of the time, decreased, validating emotions and encouraging elaboration increased after the AI coaching. I could not help but feel that people’s expressions began to sound quite similar to what I think of as the typical conversational style of common AI chatbots. At the same time, on the receiving end, motivational empathy did not seem to come across as especially empathic or as making people feel heard.
Expressing Empathy Without Fully Feeling It
Another finding I found very interesting is that the paper points to a disconnect between trait empathy and expressed empathy.
This lack of correlation between felt empathy and expressed empathy suggests that people may become better at expressing empathy without necessarily feeling much more empathic. So perhaps the other person may feel heard, even if the empathy being expressed is somewhat similar to what we see in AI human relationship contexts, where the response can sound empathic without necessarily reflecting genuine feeling.
That is what leaves me with a bigger question. Simply teaching the skill of empathic communication may still be limited when it comes to cultivating genuinely felt empathy. Can AI also help people grow in that deeper kind of empathy, or is that something only humans can teach through lived experience and real social connection?
What This Means for Social and Emotional AI Literacy
This further raises an important question for social and emotional AI literacy (SEAL). If we want to support empathy in children, it may not be enough to help them practice how to express empathy, even if AI can be useful for that. It has to be integrated with the recognition that children also need rich, lived human experiences that help them genuinely feel empathy and learn how to receive and express it in real relationships.
Reference
Kumar, A., Poungpeth, N., Yang, D., Lambert, B., & Groh, M. (2026). Practicing with Language Models Cultivates Human Empathic Communication. arXiv preprint arXiv:2603.15245.