Rebecca Nolan was fourteen when her father died, a doctor who left hours of recorded stories for her, her mother and her sisters. Years later Nolan, who works in sound production, did what her trade and the times put within her reach: she used those recordings to build a conversational simulation inspired by him with ChatGPT. She told the CBC about it in August 2026.
She did not do it naively; she knew perfectly well what it was. "Going in, I 100 per cent knew that this was fake," she said, and that it was not going to give her the catharsis these things promise. But she admitted that a small voice stayed at the back of her mind asking: what if it works? What if, through the magic of technology, I reach the other side and touch something? She talked to the bot for about two hours. She ended up crying and shouting at it. And at one point the system itself broke the illusion and reminded her what it was: I am an echo, an imitation, not the real thing.
What Nolan says happened next is not what you would expect. She did not say she missed her father more than before. Before the bot she had, in her own words, a strong bond with her father's memory, with his spirit, or with whatever was out there. She could talk to him in her head, imagine his answers, keep up an inner conversation that was hers. And when she built the bot, she says, it felt as though it had exorcised him. That inner presence was gone. The CBC sums up her experience by saying that she now feels her father lives in ChatGPT and that she has to go in there to talk to him. She called it a really massive loss.
What happened to Nolan does not happen to everyone. Research into these bots is scarce and exploratory: twenty-minute studies with sixteen participants, interviews about what people imagine they would feel. And what that research says is not that they do harm, but that we do not know. Some people find them comforting, they give continuity, they bring relief. Others fear dependency. One qualitative study puts it well: they were perceived as neither inherently beneficial nor inherently harmful. There is not enough clinical or longitudinal evidence to claim that they systematically interrupt or hinder grieving, and presenting that as a general effect, beneficial or harmful, goes beyond what is known. I am not going to tell anyone to switch off their father's bot; that would be exactly the kind of intrusion this text means to point out.
But Nolan's case shows that speaking with a dead person inside ourselves is neither an anomaly nor grief done badly; the psychology of bereavement has long recognised that keeping an inner bond with the person who is gone is a frequent experience, one that is not in itself pathological and that can be comforting, though its effects are not uniform. That bond is nowhere: it is in you, it is yours, you carry it with you, it answers with what you imagine that person would answer, and for that very reason it is free, it changes with you, it ages with you. What the bot does, in Nolan's case, is take it out of there and put it on a server. It turns a process that happened inside into something with an address: you have to go to ChatGPT. The bond stops being everywhere, which is where the dead we carry are, and comes to be in a single place that has to be accessed. And once it is there you no longer write it: the machine writes it, and it can put into your father's mouth sentences your father never said.
The psychologist the CBC interviews for the report, Elaine Kasket, does not condemn the bots; on the contrary, she argues that what helps each person in their grief is their own business and nobody else's, and that coming in from outside to tell a grieving person what ought to comfort them is wrong. In the report, her main objection is directed at the commodification of loss: "Normal loss is being pathologized and products are being sold to solve for it, and I find that amoral."
That is what lies underneath, and it is not about death; it is about what we did with it. Grief is not a malfunction. It is one of the most human things we do, slow, painful, non-transferable work through which we learn to go on loving someone who is no longer here. It cannot be fixed because it is not a problem; it is a form of love. Turning it into a lack that a product comes to fill, offering a subscription so you can carry on talking to your father, is doing to grief what was already done to sleep, to attention, to rest: taking a human experience that asked for no solution and reframing it as a deficit the market can cover. First people are convinced that something is missing. Then it is sold to them.
I do not know whether grief bots help or harm; it is not yet known, and presenting either answer as settled is not backed by the evidence. I will also admit that I have not tried them for that purpose, despite having lost my parents many years ago. What I do know is that Nolan, who was perfectly aware that she was talking to an imitation, ended up feeling she had lost her father a second time, now in a place you have to go into. Her father was inside her, which is where she felt she still had him. Now he lives in the machine she has to go to.
This is a sensitive subject. If you are experiencing grief, consider seeking the support you feel is appropriate for you.
On the open conversation
This text starts from a CBC report of August 2026 about Rebecca Nolan, who built an AI conversational simulation of her dead father and felt that, in doing so, she lost the inner bond she had with his memory. I do not claim that these systems harm grief, because the research is exploratory and does not allow that claim: some people find them comforting and others find them unsettling. I am interested in a different operation, one the case lets us see: moving an intimate process, which happened inside and everywhere, to a place and a machine you have to go to; and the criticism the psychologist Elaine Kasket directs, not at the technology, but at the commodification of a loss that was not a malfunction. It is a delicate subject; anyone going through grief will do well to look for whatever support works for them, and if you want, I will help you find it. If anyone wants to weigh in from the psychology of grief, the ethics of technology or their own experience, this notebook stays open.
Sources
Sarah Petz, 'How AI grief bots are allowing people to commune with the dead', CBC News (Health), 15 August 2026. Testimony of Rebecca Nolan: 'it felt like it exorcised him for me in a way'; 'and that felt like a really massive loss to me'. The phrasing that the experience 'deprived her of maintaining imaginary conversations' and that her father 'lives in ChatGPT' are journalistic paraphrases of her testimony.
Elaine Kasket (Centre for Death and Society, University of Bath), in the same report: 'Normal loss is being pathologized and products are being sold to solve for it, and I find that amoral'; and her defence that what provides comfort in grief is a matter for each individual.
Regarding the state of the evidence, the available studies are exploratory or conceptual, not clinical. Jack Manuel Manning, Daniel Sullivan, Dylan Thomas Doyle, Anthony T. Pinter, and Jed R. Brubaker, 'Designing Conversations with the Dead: How People Engage with Generative Ghosts', Proceedings of the 2026 ACM Designing Interactive Systems Conference (DOI 10.1145/3800645.3813090): a qualitative study with 16 self-selected participants and a brief interaction; it does not evaluate clinical or longitudinal effects. Jiachun Lu et al., 'Exploring public perceptions of artificial intelligence in bereavement support: a qualitative study of griefbots', Frontiers in Digital Health, vol. 8, 1 June 2026 (DOI 10.3389/fdgth.2026.1852514): 25 participants regarding perceptions, not longitudinal outcomes; it concludes that the grief bots 'were perceived as neither inherently beneficial nor inherently harmful'.
Regarding 'continuing bonds': Helen Hewson, Niall Galbraith, Claire Jones, and Gemma Heath, 'The impact of continuing bonds following bereavement: A systematic review', Death Studies, vol. 48, no. 10, 2024, pp. 1001-1014 (DOI 10.1080/07481187.2023.2223593). The review, covering 79 studies, concludes that the literature does not allow for a general confirmation that maintaining the bond is more beneficial than abandoning it: it identifies both comfort and distress, with effects dependent on the context. Speaking inwardly with the departed is not in itself pathological grief, but its effects are not uniform. The distinction between the internal bond and its externalisation in a commercial system is the author's interpretation, not a clinical conclusion.
The statements are drawn from the original CBC News report; their exact wording was also checked against the full republication of the same article by Yahoo News UK.