Grieving people prefer bots that speak like their deceased loved ones, says study

A new study from the University of Colorado Boulder investigated how people emotionally engage with “generative ghosts” – AI chatbots trained to simulate deceased loved ones.

The study found that users generally prefer bots that speak in the first person of the deceased (‘Reincarnation’) to bots that simply describe the deceased (‘Representation’), despite being aware of the risks of being too attached.

Lead author Jack Manuel Manning, a doctoral candidate in information science, presented the research at the Association for Computing Machinery’s Interactive Systems Design Conference in June.

Connection app that anyone can use

adult friend finder

— Casual Connections Selected by Readers

crater

— Top picks for finding a date

hinge

— A popular choice for regular meet-ups

You can purchase products through affiliate links. If you buy something through links on our site, Mashable may earn an affiliate commission.

Researchers conducted a qualitative study with 16 participants between the ages of 22 and 50. Each participant interacted with two versions of the chatbot of their deceased loved one. One spoke in the first person, directly imitating the deceased, and the other described the person in the third person, like a narrator.

The study found that participants found the first-person “reincarnation” mode more emotionally vivid and pleasant, but some worried about the psychological risks of relying too much on it. One participant, referred to as P4, stated that the experience had an unexpected outcome. “in [reincarnation]I feel like I’m getting the closure that I really needed,” she told researchers.

Mashable Trend Report

Another participant, P11, expressed further alarm: “I’m worried that over time they’ll become dependent on the voice…It’s going to end up very similar to people falling in love with an AI character.”

The researchers also found that participants cared more about whether the chatbot’s tone and phrasing felt emotionally “right” than whether it was factually accurate.

In some cases, participants continued to treat the third-person “expressive” chatbot as if it were their loved one speaking directly to them, effectively ignoring the distinctions that researchers had built into the study design. One participant, P12, explained, “I don’t see this chatbot as a human, but it still says ‘you.’” I think it’s just thinking about what question to ask the person and confusing that with asking the chatbot a question. ”

The study’s authors were upfront about its limitations, particularly the small sample size. With only 16 participants, the researchers acknowledged that their findings did not capture “the full range of cultural, religious, and individual perspectives on grief and technology,” and noted that mourning practices vary widely between communities and can influence how people respond to these tools.

This study also focused on single short-term sessions rather than repeated or long-term use, leaving open questions about how attachment builds or fades over time.

The paper argues that future developments of these systems must weigh the psychological benefits against the risks of unhealthy dependence, and calls for careful consideration of consent and family governance before introducing such tools to bereaved users.