A new study from researchers at the University of Colorado Boulder reveals that people find comfort in AI chatbots designed to mimic deceased loved ones, even when the bots make factual errors about the person's life or personality.
The research, published recently, involved a small group of mourners who interacted with chatbots trained on memories and details of people who had died. Participants reported feeling emotionally supported by conversations with these digital recreations, regardless of accuracy lapses. The findings suggest that emotional resonance matters more than perfect fidelity when it comes to whether people accept these AI versions as meaningful.
Lead researcher Lia Kvavilashvili and her team found that mourners prioritized feeling understood and emotionally validated over receiving historically accurate information. When a chatbot captured the tone, warmth, or personality style of the deceased, participants felt the interaction was "real" in ways that mattered to them psychologically. Occasional factual mistakes about specific memories or details did not substantially diminish this sense of connection.
The study raises both possibilities and concerns. For people struggling with acute grief, an AI companion offering emotional support without judgment could provide a useful supplement to traditional grief counseling. However, the findings also highlight risks. Relying on factually inaccurate representations of the dead could distort memories or prevent people from processing grief through other, more grounded approaches.
Researchers emphasize that their sample was limited and that the long-term psychological effects remain unknown. Some participants reported that interactions with the bots felt uncanny or unsettling, suggesting this technology does not work uniformly for everyone.
The work sits at the intersection of artificial intelligence, psychology, and ethics. As AI becomes increasingly sophisticated at mimicking human conversation, questions about appropriate use in grief contexts grow more pressing. The researchers call for clearer guidelines around disclosure. Users should know they are talking to algorithms, not genuine digital