First study on AI ghosts reveals people prefer talking to the dead as reincarnations, not narrators

People are talking to AI ghosts to cope with grief - does it actually help?

Artificial intelligence can now recreate the voice and personality of someone who has died, turning old messages, photos, and recordings into a chatbot that carries on a conversation. Whether that comforts grieving people or makes loss harder has remained largely an open question. A new study found that most participants preferred AI companions that spoke as their deceased loved one rather than simply describing them. While many found the experience comforting, the researchers also identified ways these digital memorials could become emotionally risky. Speaking as the dead Researchers at the University of Colorado Boulder (CU Boulder) recruited 16 people, aged 22 to 50, who had each lost a close relative or friend. In two separate 20-minute chats, every participant talked with two AI versions of that person, and then compared them. The team calls these systems generative ghosts, AI systems trained on data about a deceased person. One version described the person in the third person, as a knowledgeable narrator would. The other spoke as them, in the first person, as though the dead had rejoined the conversation. Jack Manning, a PhD candidate in information science who took up the subject after losing his sister, expected the sessions to feel disturbing. "I ended up being completely wrong. People thought it was amazing," said Manning. How users experience AI ghosts The distinction between speaking as the dead and speaking about them comes from an earlier framework co-authored by Professor Jed Brubaker, who runs the lab. "To our knowledge, we are conducting the first user experience studies of simulated AI ghosts," said Brubaker. The ghosts were not fully automatic. Out of view, a second researcher fed each person's details into a chatbot and passed along its edited replies, which took about 60 to 90 seconds each. Everyone leaned to the speaking-as version. "I can see her. I can feel her," said one woman in her early 30s, after hearing the AI speak in the voice of her late grandmother. For a 50-year-old woman, the effect reached back into childhood. Her grandmother had once promised to visit her after death, and she read the exchange as that promise kept. "Thank you for following through on your promise to visit me, it was so so powerful," she typed. When the illusion disappears The clearest lesson had almost nothing to do with getting facts right. Participants forgave the AI when it invented details or muddled events. What they could not overlook was language that sounded nothing like the person. One young man nearly walked out when the ghost of his stepfather called him "champ," a nickname the man would never have used. "Once it sounded off, I just couldn't get back into it," he said, and after that the illusion disappeared. That pattern held across the sessions. A single wrong endearment, or a phrase in the wrong language, could break the spell while every surrounding fact stayed correct. Authenticity over accuracy Earlier research on griefbots leaned on accuracy and detail, but here emotional authenticity mattered far more. People noticed how the conversation moved, too. One woman said the replies felt more like email than texting, too long and too dry, without the quick back-and-forth she remembered. Shorter messages, pauses and the odd emoji, she suggested, would have made it feel true. Blurring the line Something strange happened whenever the AI spoke in the third person. People replied as if the person stood right in front of them. One grandson, told plainly that the system only described his late grandfather, still asked him directly whether he was on the right path in life. Emotional instinct kept overriding the design. Watching the logs, the team saw participants steer each exchange toward connection no matter which mode was running. The illusion depended partly on the griever's willingness to participate in it. The same line could land very differently from one person to the next. One woman preferred the distant, third-person version because her grandmother really had talked at people rather than with them. The cooler tone rang true. How real a ghost felt depended less on the technology than on the unique relationship the participant had with the deceased. Comfort and its cost Every participant said they would use the system again. Almost as fast, most of them named a fear. The comfort felt strong enough that coming back to it could be hard to resist. Most participants worried more about how the technology might affect someone else than themselves. One woman found her own conversation healing, yet dreaded what the same tool might do to her still-grieving brother. She worried he could never stop pressing it for answers he would never accept. Grief researchers have long held that healthy mourning can mean keeping a lasting bond with the dead rather than severing it, an idea called continuing bonds. The double edge of AI Letting go is not the only path. A digital afterlife industry is now betting that software can carry that bond forward. That possibility concerns Manning most. He thinks about his eleven-year-old self, and how an AI answering late at night as his dead sister, with no adult nearby, could have gone badly. What the study makes clear is that a convincing ghost depends more on emotional accuracy than factual accuracy. One system can soothe one grieving person while trapping another. Manning argues these tools should be developed cautiously before they become widely marketed to grieving families. For all his caution, he has come to see the promise as well as the peril. "It can also be an incredibly meaningful experience for people that enables them to get some closure and peace," said Manning. The study is published in the journal Proceedings of the 2026 Designing Interactive Systems Conference. -- - Like what you read? Subscribe to our newsletter for engaging articles, exclusive content, and the latest updates. Check us out on EarthSnap, a free app brought to you by Eric Ralls and Earth.com.

If you could chat with an AI ghost, what would you want them to say? New study explores

Inside a second-floor lab at CU Boulder, Jack Manning and Jed Brubaker are resurrecting the dead. As the researchers look on, study volunteers log in to Zoom and chat with AI-generated representations of lost grandparents, siblings, parents and family friends. Some get emotional: "I can see her. I can feel her," said one 32-year-old woman during a text-based conversation with her grandmother, who died five years ago. "It just feels like I'm getting the closure I needed." Others plan the next visit: "It was so so powerful," typed a 50-year-old woman to the ghost of her beloved. "I'd like for you to come to me again." These interactions, chronicled in a paper published in the Proceedings of the 2026 Designing Interactive Systems Conference, offer the first scientific glimpse of how people use "generative ghosts" -- the increasingly popular AI agents trained on data about the dead. Among other things, the study found that participants preferred ghosts that spoke in the first person, acting as a resurrection rather than a representative. The more accurate and lifelike the ghost's emotional tone, dialect and conversational rhythm, the better. "We originally thought it might feel very Black Mirror creepy to people and make them uncomfortable," said first author Manning, a Ph.D. candidate in information science who found his way to the unusual field of study after losing his sister. "I ended up being completely wrong. People thought it was amazing." Conjuring the dead with AI Once viewed as science fiction, generative ghosts, sometimes called "griefbots" or "deathbots," are fast becoming a commercial reality. Platforms like Project December and Séance AI use journal entries, social media posts and texts from the deceased to train text-based ghosts for surviving loved ones to chat with. Others, like HereAfterAI, invite users to submit voice recordings and photos of themselves to create multimedia ghosts for their loved ones to interact with after they're gone. Some startups have even created fully immersive, virtual reality options, enabling grieving clients to literally walk with a hologram of the dead. Brubaker, an associate professor of information science, predicts that generative ghosts will soon be a regular part of life as we know it. But given their potential to both help and harm, he believes they should be designed with solid research as a guide. That's where his lab comes in. "To our knowledge, we are conducting the first user experience studies of simulated AI ghosts," Brubaker said. Don't call me 'champ' For their inaugural study, the research team recruited 16 people, ages 22 to 50, who had lost a close relative or friend. Participants logged in to Zoom for a brief on-camera interview with a facilitator, who gathered biographical information and other details about the deceased. Meanwhile, in the background, a second researcher plugged this information into a large language model (LLM), building a ghost in real time. Participants chatted with two iterations, each for about 20 minutes. One ghost spoke in first person ("I remember going to the beach together"). Another used third person ("She loved going to the beach with you"). The facilitator stood by to intervene if things got uncomfortable, while the operator behind the scenes used the conversation details to fine-tune the ghost. Subsequent interviews found that, across the board, participants preferred the "reincarnation" over the "representation." While people seemed willing to overlook occasional inaccuracies, or "hallucinations," spun up by AI, they cringed if the bot used the wrong term of endearment. (When the ghost of one participant's stepfather called him "champ" -- a term he would never have used -- the participant nearly called off the session.) Users also preferred shorter sentences with emojis rather than the rambling paragraphs that AI tends to produce. Perhaps the most interesting finding came at the end, when participants were asked if they would use the technology again. Surprisingly, everyone said yes. But almost all added that they feared what would happen to their grieving loved ones if they got their hands on one. Why more research is needed Manning lost his sister to a heart condition when they were kids and had longed for years for more meaningful ways to memorialize her. When he heard about AI ghosts, he was initially horrified -- which is exactly why he felt he was a good fit to study them. "I felt it was important for me to do the work because the people who are the largest fans might skip the empirical research and just make a product," Manning said. The lab has already begun the next studies, including one working with mental health professionals to analyze the benefits and risks of interacting with AI ghosts. In addition to seeing their potential peril, Manning said he now also recognizes their promise. "I think a lot about 11-year-old me. If I had access to ChatGPT and it started responding as my sister late at night without supervision ... that is a very scary thought," he said. "But as we have learned through this paper, it can also be an incredibly meaningful experience for people that enables them to get some closure and peace."

Study finds humans will talk to AI ghosts of the dead as reincarnations, and it's pretty grim

The first AI ghost study is in. The results are about as complicated as you'd expect. A new study from the University of Colorado Boulder confirms something that sounds both impressive and concerning. People find interacting with AI simulations of their dead loved ones deeply meaningful, and most will come away wanting to do it again. The researchers call it a "generative ghost," which is a clear reference to generative AI, but I'd still prefer to call it unsettling. So what did the study actually find? Doctoral candidate Jack Manning and associate professor Jed Brubaker recruited 16 participants aged 22 to 50, all of whom had lost someone close to them. Recommended Videos During individual Zoom sessions, a second researcher quietly used an LLM to build a ghost of the deceased (in real time) from details provided by the participant, an AI-based reincarnation, if you will. Each participant chatted with two versions of the generative ghost: one that spoke in first person ("I remember going to the beach together") and one that used third person ("She loved going to the beach with you," where you is the participant). Participants unanimously preferred the first-person "reincarnation" over the third-person "representative," which, I'll admit, is the part I find most unsettling. So who is building these, and why does it matter? Small factual inaccuracies were forgiven during the interaction. However, wrong terms of endearment were not. For instance, when one stepfather's ghost called his stepson "champ," a word he'd never used, the participant nearly ended the session. This is the first user experience research on AI ghosts, published by the Association for Computing Machinery (via CU Boulder). And if you don't already know, commercial services like Project December and HereAfterAI are already selling AI ghosts as a product. The study's own participants flagged a significant concern. While everyone said they'd use a ghost again, almost all worried people who've lost their loved ones would become addicted to one. The lab has already initiated a follow-up study with mental health professionals to assess the psychological benefits and risks of generative ghost interactions.