Can a machine comfort you? Research by the University of Manchester and Durham University, published in the journal Emotion, compared emotional support messages written by three large language models, ChatGPT-4o, Claude 3.5 Sonnet and Gemini 1.5 Pro, with messages written by people. Across five studies involving 1,233 participants, the scenarios were everyday and non-clinical, and participants were not told whether a message came from a human or an AI.
The language models largely mirrored the strategies humans use to manage others' emotions, with ChatGPT the closest match. In anger and fear scenarios, AI-written messages were actually rated more emotionally supportive than human ones and left recipients feeling more positive; for sadness, humans and machines performed equally well. The advantage was not explained by simple validation of feelings. Instead, specific practical suggestions, such as breathing techniques or step-by-step reframing of a problem, improved outcomes regardless of who wrote them.
Notably, many participants correctly guessed which messages were machine-written, yet this knowledge did not reduce the rated benefit. The authors stop short of recommending chatbots as counsellors, but the findings suggest that well-designed AI support may have a role in everyday emotional life, and that what makes support effective, human or artificial, is concrete help as much as warmth.