AI is becoming the shoulder we cry on, and that worries me
I expected artificial intelligence to be faster in performing calculations or data analysis than humans. But emotional support is one of the last things I could imagine it being better at. New research led by the University of Manchester, alongside Durham University, found that leading AI chatbots could match or sometimes outperform humans when responding to everyday emotional problems. The researchers conducted five studies involving 1,233 participants and tested responses from ChatGPT-4o, Claude 3.5 Sonnet, and Gemini 1.5 Pro.
The researchers conducted five studies involving 1,233 participants and tested responses from ChatGPT-4o, Claude 3.5 Sonnet, and Gemini 1.5 Pro. Participants evaluated messages responding to situations such as being passed over for a promotion or returning to an untidy shared home. They were initially kept unaware of whether each response came from a person or an AI system.
AI was particularly good at handling anger and fear
Responses generated by the chatbots were rated as more emotionally supportive in scenarios involving anger and fear. These even helped participants convey more positive emotions like calmness and happiness. Human and AI responses performed similarly in situations involving sadness though. Surprisingly, simple validation did not explain the advantage. Telling someone that their feelings were understandable made little difference by itself.
Where the study found the strongest impact was when these responses carried actionable support. Something along the lines of breathing exercises, practical next steps, or even structured ways to reconsider a situation. AI systems tended to provide this kind of detailed guidance more consistently than the human participants. The findings also show how people could become better supporters. A useful response often requires something more concrete than a sympathetic sentence.
A chatbot’s support ends with the message
The experiments focused on brief, written responses to everyday situations. So they did not examine therapy, other kinds of emergencies or prolonged distress. It also didn’t touch on the consequences of relying on a chatbot over months or years. The researchers also explicitly caution against treating AI as a replacement for human support.
Durham University researcher Dr Sarah Walker highlighted the distinction particularly well. A chatbot can recommend taking a walk, while a person can join you on that walk, being present, and continuing to help even after the conversation ends.
