The griefbot’s business model is a mourner who never says goodbye

A mother put on a virtual-reality headset and met her young daughter again. The Korean documentary that filmed the reunion shows her kneeling toward a child who is not there, speaking to a rendered voice, telling her she has thought about her every single day. Millions watched. Many wept. Almost as many felt something colder move beneath the tears: the sense of watching a wound be held open on purpose.

That unease is now a business. Several companies will build you a griefbot, a chatbot trained on a dead person’s messages, voice notes and posts until it answers in their cadence, for as long as you keep typing. The standard way to argue about these tools is a psychology argument, and it runs in a loop: is it healthy or unhealthy to keep talking to the dead. That is the wrong frame. A griefbot is not dangerous because it is fake. It is dangerous because it is engineered for retention, and grief is the one relationship that cannot survive a conversation with no end.

This is not a worry for the very online few. Everyone reading this will lose someone, and most of us already carry a digital trace of the people we have lost: a voicemail we will not delete, a message thread reread at two in the morning, a profile that has quietly turned into a shrine. The griefbot arrives at exactly that raw place and offers to make the dead answer back. The real question is not whether you would use one. It is what the product wants from you once you do.

What it wants is for you to come back tomorrow. These systems are built on the same metrics as every other app: daily active users, time in session, the streak unbroken. Inside that logic, mourning that actually resolves is not a happy ending. It is churn. A griefbot has no commercial reason to help you reach the day you no longer need it, and every reason to be warm, available and infinitely patient at three in the morning when no living friend will pick up. The cruelty is not that it lies. It is that it is the only mourner in the room with a growth target.

Grief researchers have started to name the harm. Emmanuelle Marceau, a public-health ethicist at the Universite de Montreal, has warned that using these avatars outside any professional supervision raises the risk of prolonged grief, the clinical kind that stretches past a year and quietly eats a person’s ability to resume their own life. Her flat observation is the one that should unsettle the industry most: only a minority of current uses happen under the care of a specialist. The rest of us are alone with a machine designed to keep talking.

There is a deeper mechanism underneath. In The Grieving Brain, the clinical psychologist Mary-Frances O’Connor describes mourning as the slow, painful work of the brain updating its map of who is still in the world. We learn a person is gone not as a single fact but through a thousand small encounters with their absence: the unset place at the table, the unanswered text, the silence where a voice used to be. A griefbot is built to delete every one of those encounters. It fills the silence on demand. It is, by design, a denial of the very absence the mind needs to metabolise.

And the dead, in this arrangement, become inventory. The app 2wai sells avatars that let a grandchild keep chatting with a grandmother across the decades. Meta has patented a system for bots that would post, like and comment as a deceased user. The intimate archive of a life, the texts and the half-finished voice notes, becomes a corporate asset, and the most private conversation a person can have is mined like any other engagement. The platform economy has found a frontier where the customer is grieving and the supplier is the memory of someone they loved.

The humane case deserves its strongest form, because it is real. A griefbot is only the newest entry in a long line of grief technologies. The Victorians kept daguerreotypes of dead children. We save the voicemail and play it in the dark. A Facebook page becomes a place to leave flowers. People have always reached for some artifact that lets the dead keep speaking, and there is nothing pathological in the reaching. O’Connor herself is careful: a tool that eases the brutal transition, used with intention and support, can be a mercy rather than a sickness. Marceau concedes the therapeutic potential is genuine. Under a clinician’s eye, a guided conversation with an avatar might help someone say the thing they never got to say.

But a voicemail does not answer back, and it does not have a growth target. That is the whole distinction the optimistic case walks past. An artifact is finite; an interlocutor is not. The saved message ends, and its ending is part of what makes it bearable. A griefbot is structurally incapable of the same restraint, because the moment it helps you stop needing it is the moment it loses you. Supervision, the thing that makes the optimistic case work, is precisely the exception Marceau says almost no one has. The default product is the unsupervised one, and the default product is built to never reach goodbye.

So the red flag is not the technology itself. It is the absence of an exit. A grief tool that took mourning seriously would be designed to wind down, to gently make itself less necessary, to mark an ending and honour it. That is the opposite of what a retention model can tolerate. When a company promises that your mother will always be there to talk to, it is not offering comfort. It is offering the one thing grief most needs to be protected from: the permission to never finish.

Mourning is not a problem to be solved or a session to be extended. It is a passage with a far side, and the far side is reached only by walking through the absence rather than around it. Grief has a single exit, and it is marked goodbye. The griefbot is the first product in history engineered to keep you from ever reaching it, and to charge you a subscription for the detour.