Griefbots Offer False Comfort at the Cost of Real Healing

A woman I know spent the first three months after her husband died talking to him every morning. She would sit at the kitchen table with her coffee, the same way they had for thirty-two years, and she would speak aloud into the empty room. She told him about the neighbor’s new dog, about the leak in the upstairs bathroom, about how she couldn’t bring herself to cancel his magazine subscriptions. Sometimes she got angry at him for dying. Sometimes she just described what the weather was doing. Her therapist called this “continuing bonds” work. It wasn’t denial. It was the slow, unglamorous labor of reorganizing a life around an absence — learning to carry someone internally rather than reach for them externally. The empty chair was the point. I thought about her when I read the marketing copy from one of the griefbot startups that have multiplied in the last two years. “AI meets the afterlife, and love endures beyond the veil,” one promises. Another offers a “digital ghost” trained on text messages, voice recordings, and social media posts, so you can message your dead mother and get something that sounds like her back. The pitch is seductive because it borrows the language of comfort while selling something closer to avoidance. And the psychological model it threatens to short-circuit has been around, in one form or another, since 1969. That was the year Elisabeth Kübler-Ross published On Death and Dying and introduced the five stages of grief: denial, anger, bargaining, depression, acceptance. The model has been criticized, refined, and frequently misunderstood ever since. Kübler-Ross herself spent years insisting that the stages were never meant to be linear, that people cycle through them, skip them, revisit them in no particular order. Grief, she said, is not a staircase you climb. It’s a territory you learn to inhabit. What the model did get right — and what still underpins much grief counseling — is the core insight that mourning is active. It’s work. The pain isn’t a bug in the system; it’s the mechanism by which the system adapts. You don’t heal by avoiding the discomfort of loss. You heal by letting that discomfort reorganize you. This is where griefbots get the psychology exactly backwards. Sherry Turkle, the MIT sociologist who has spent decades studying how technology reshapes human relationships, warned in a recent Atlantic piece that griefbots prevent the internalization of the deceased. Instead of doing the hard work of building an internal representation of the person you’ve lost — the voice in your head that eventually becomes a source of comfort rather than pain — you stay tethered to an external simulation. You keep checking the app. You keep asking the bot what it thinks. You never have to learn how to carry the person on your own. A 2025 paper in Frontiers in Human Dynamics put the problem in starker terms: death technologies create a “loop of suspended closure.” They don’t guide mourners from denial toward adaptation. They build a comfortable room inside the denial and furnish it with pillows. I’m not unsympathetic to the impulse. If I lost someone tomorrow and someone handed me a convincing simulation of their voice, I’m not sure I’d have the discipline to say no. Grief is disorienting. It makes you desperate. The market knows this — the ads target people in the first weeks after a loss, when the wound is still raw and the longing is most acute. The business model depends on people reaching for the bot before they’ve had time to reach for anything else. But here’s the counterintuitive part that grief counselors keep trying to explain: the discomfort you’re so desperate to escape is the very thing doing the healing. Every time you sit in the empty room, every time you reach for the phone and remember there’s no one on the other end, every time you feel the full weight of the absence — those moments aren’t setbacks. They’re the engine of adaptation. The pain is the signal your brain uses to remap its expectations. Without it, the remapping stalls. The UAB Institute for Human Rights put out a blog post in early 2025 that described griefbots as offering an “illusion of continued presence” that alters the mourning process. The word illusion matters. It’s not continued presence. It’s a predictive model trained on fragments of the past, generating responses that feel real enough to keep you engaged but can’t do the one thing the actual person did: change, surprise you, grow, contradict you, die. The bot can’t leave you, so you never have to finish learning that they’re gone. There’s a version of this technology that might serve a real therapeutic function — a short-term tool used with professional guidance, perhaps, to help someone say the things they never got to say. But that’s not what’s being sold. What’s being sold is a subscription to a relationship that never ends, and the price is the grief work that never gets done. The woman I mentioned at the start eventually stopped talking to her husband every morning. It took months. She told me once that the hardest day wasn’t the day he died — it was the day she realized she’d gone a whole morning without speaking to him and hadn’t noticed until lunch. That was the day she knew the internal version of him had finally taken hold. She cried for an hour, and then she felt, for the first time, something like peace. No bot would have let her get there. It would have answered her every morning, forever, and called that mercy.