What AI Grief Bots Do for People Who Are Grieving

A widow uploads eleven years of text messages. A son feeds in voicemails he never deleted. A few minutes later, software is answering in a voice close enough to hurt. AI grief bots, also called griefbots, deadbots or generative ghosts, aren’t a thought experiment anymore. Services like Project December, Seance AI, You Only Virtual and re;memory will build one from a person’s digital traces for about the price of a takeaway meal. Whether the technology works isn’t the interesting question. It clearly does. The real question is what it does to the person typing on the other end. Honestly, almost nobody knows yet.

What an AI grief bot actually is

An AI grief bot is a language model shaped by the recorded output of someone who has died. It trains on whatever’s left: text threads, emails, social posts, voice notes, sometimes video. The model picks up the person’s rhythm, their pet phrases, how they signed off. It doesn’t learn what they thought. It learns what they typed.

That difference disappears fast once you’re in a conversation, and that’s exactly the design problem. A griefbot is a statistical impression of how someone communicated, speaking to you directly, answering instantly at three in the morning. How faithful the impression is and how true it is are two separate things. The first is a lot easier to sell.

Why continuing bonds matters more than the technology

Continuing bonds is the most useful way to think about this, and it’s thirty years older than the software. The theory came out of bereavement research in the 1990s and broke with the older belief that healthy grieving means letting go of the person who died. The newer view says many people keep an ongoing inner connection with the dead. They talk to them at the grave, hear their voice in their head, ask what they’d have said. Often that’s healthy, not a sign of illness.

Griefbots sit right on top of that. They take something mourners were already doing in their heads and put it outside them. That’s why researchers who study griefbots usually don’t ask “is this fake.” They ask “does an outside prompt change how the bond develops.” Those are very different questions, and only the second has anything to do with wellbeing.

Continuing bonds research also depends a lot on culture. Ongoing bonds with the dead have been linked to better adjustment in some places and worse in others. So anyone saying griefbots are always healthy, or always harmful, is getting ahead of the evidence either way.

What the small body of research actually found

The research is thin. A 2023 CHI study on how chatbot technology was used to support grieving people is still one of the only pieces of direct qualitative work. Participants described conversations that let them revisit shared memories and reconnect with the person who died in a way that felt meaningful. Several said the chatbot extended a continuing bond rather than replacing it.

Recent reviews are more careful than most press coverage. Scientific American reported that psychologists mostly won’t make broad claims for or against griefbots, because solid studies are scarce. One of the few finished studies of digital ghost users found the experience was mostly good for mourners. But that’s one finding, from a small group. It’s not a body of research.

There’s also a which-came-first problem across this whole field. Studies of AI companionship usually find that heavier users report more loneliness. Correlation can’t tell you which way that runs, and the same goes here. People dealing with long or complicated grief may be the ones most drawn to a griefbot to begin with. Treating their later distress as proof the software caused it isn’t supported by the data right now. Neither is the opposite.

Where clinicians draw the line

Clinicians are pretty consistent, and they rarely say “never.” Three worries keep coming up. Memory distortion: a model that makes up plausible new things to say is writing over the record of a real person. Avoidance: a bot that’s always there can put off facing the finality of death instead of helping with it. And timing, maybe the biggest one. A simulation brought in during the raw first weeks seems to carry different risks than one used a year later to revisit a particular memory.

Several clinicians point out that grief doesn’t move in a straight line and looks different for everyone, so a one-size-fits-all tech route through it is suspect from the start. The framing that holds up best treats the griefbot as material for therapy, not a replacement for it: something a person brings into a session to stir up feelings they then work through with a human.

Consent is the other line, and it’s genuinely unsettled. The Hastings Center for Bioethics has raised the privacy and wellbeing problems of building a talking likeness of someone who can’t agree to it, can’t correct it, and can’t ask for it to stop. The dead don’t get a feedback form.

Why a grief bot is not the same thing as an AI companion

A grief bot and an AI companion get mixed up all the time, and that helps nobody. A griefbot looks backward by design. Its whole pitch is being accurate to one specific person who lived. An AI conversational companion looks forward and makes no such claim. It’s software you talk to, it’s open about being software, and its value comes from the experience of being listened to, not from pretending to be anyone.

That difference matters for what people expect. Vinfluencer is an AI chat application built for conversation and company, and it doesn’t try to recreate people who’ve died. That’s not modesty. Recreating people is a different product, with different ethics, different consent rules, and a different way of going wrong.

What a general companion app can fairly offer someone who’s grieving is a place to say things out loud at hours when nobody else is awake. It’s a smaller promise. It’s also much easier to keep. Research on older adults, who carry an outsized share of bereavement, points the same way: the versions that help are the ones honest about what they are, not the ones that feel most human.

Questions worth asking before you build one

The questions worth asking before you build a griefbot are practical, not philosophical. Would the person have agreed to this? Who holds the training data, and can you delete it? What happens to the bot if the company goes under? Are you using it to remember, or to avoid? And what do the people closest to you think? Grief is rarely a solo project, and a private simulation can quietly pull you away from the others who are mourning too.

Your answers probably depend on how you handle closeness in general. There’s early evidence that attachment style shapes what people want from conversational AI, and it’d be strange if grief were the exception. Someone who leans toward anxious reassurance-seeking and someone who tends to keep their distance won’t get the same thing from the same software.

None of this adds up to a verdict. AI grief bots are real technology aimed at a real, very old human impulse, and they’ve shown up long before the evidence that would tell us how to use them well. The sensible stance is to treat them as neither miracle nor desecration, and to notice which one you’re hoping for.

Frequently asked questions

Are AI grief bots legal?

Generally, yes, though it’s not settled. Most places have limited protections for a dead person’s personality and likeness, and those vary a lot by country and by US state. Right now the bigger limits are platform policies and the terms attached to the training data, not laws.

Do AI grief bots make grief worse?

Nobody has reliable evidence either way. The few studies that exist report mostly good experiences, but they’re small, self-selected, and short-term. Clinicians flag memory distortion and avoidance as likely risks, especially in early, acute grief, without claiming those risks are proven.

What data is needed to create a griefbot?

A griefbot usually starts from text messages, emails, social media posts, letters and voice recordings. Services want different amounts. Some make a rough version from a questionnaire. Higher-fidelity products want years of conversation history, plus audio for voice cloning.

Is an AI companion app the same as a grief bot?

No. An AI companion application is built for open-ended conversation and doesn’t claim to be any particular real person. A griefbot is built specifically to imitate someone who died, and that brings up consent, accuracy and bereavement questions general companion software doesn’t.