quiddity
Griefbots and Voice Preservation: What Families Should Know
What griefbots are, what early research says, and how an archive that only speaks a person's own recorded words differs from one that makes up new ones.
The grief impulse is ancient and entirely human: we want, against all reason, one more conversation. One more story. The sound of a voice saying something new.
Griefbots — AI systems built from a person’s messages, emails and recordings so that people can keep talking with a simulation of them after they die — offer a version of that. The technology is real, easy to reach, and growing. So are the questions about it.
This piece tries to give you a clear picture: what griefbots are, what the early research suggests, and where the ethical lines fall. Because we make a voice product ourselves, it also explains exactly how Quiddity’s Avatar works, so you can judge it against the same lines. That matters whether you are thinking about someone you have lost, someone you are caring for, or your own voice.
This article is for information only and is not a substitute for grief counseling or professional mental health support. If you or someone you love is struggling with a loss, please reach out to a qualified professional. If you need someone to talk to right now, in the United States or Canada you can call or text 988, free and confidential, 24 hours a day: the 988 Suicide & Crisis Lifeline in the US, or 9-8-8: Suicide Crisis Helpline in Canada. Anywhere else, find a free, confidential helpline near you.
What is a griefbot?
A griefbot is an AI system trained on a person’s digital footprint — their texts, emails, social posts and voice recordings — so that living people can hold new conversations with a simulation of them. They are also called “deadbots,” “digital ghosts,” or “AI afterlife” services. The defining feature is that they generate replies the person never gave, based on patterns learned from what they left behind.
Some services were built for grief. Others found the use arriving on its own, as bereaved people pointed general-purpose AI companions at someone they had lost. The field grew fast enough that researchers at the University of Cambridge’s Leverhulme Centre for the Future of Intelligence called for safeguards in 2024, warning that current design practices left people open to psychological harm and “unwanted digital hauntings” (University of Cambridge, May 2024).
What does the research say so far?
Honestly: not much yet, and what exists is mixed.
A small pilot study with 10 participants who used AI simulations of deceased loved ones, reported by Scientific American in November 2025, found that participants described real comfort, and some became more socially engaged rather than less. Ten people is far too few to generalize from. It is an early signal, not evidence.
The risks are real too. An estimated 7 to 10 percent of bereaved adults may develop prolonged grief disorder, a condition in which grief does not ease with time (Newsweek, 2026). Whether an ongoing simulation of the person who died makes that more likely for vulnerable people is still an open question.
Did the person ever agree to this?
This is the sharpest question griefbots raise. Often the answer is unclear, or no. A simulation can be built by a family member from messages the person wrote for entirely different reasons, without anyone ever asking whether they wanted to be recreated.
The Cambridge researchers put consent at the center of their recommendations. They suggested services prompt people to consider how the person wanted to be remembered, and build in a “dignified retirement” — a way to bring a simulation to a meaningful close rather than leaving it running indefinitely. They also noted that when a service contract was signed before death, surviving family may have no power to pause a simulation, even when it has become distressing (University of Cambridge, May 2024).
Two different things that both use a voice
It is easy to put every “AI voice” product in one bucket. There are really two different designs, and the difference is not whether a synthetic voice is involved. It is whose words it says.
| A generative griefbot | A verbatim voice archive | |
|---|---|---|
| What it says | New sentences a model composes | Only words the person actually recorded |
| When it is made | Usually after death, from data left behind | While the person is alive, by the person |
| Who decides what it holds | Often the survivors | The person, story by story |
| When it has no answer | It produces a plausible one | It says the person never talked about that |
| The voice you hear | Synthetic, saying things the person never said | The original recording, or a synthetic voice saying only the person’s own words |
Both can sound like the person. Only one of them puts new words in their mouth.
How Quiddity’s Avatar works, plainly
Quiddity is built on the second design. We think you should know exactly what that means, including the parts that are synthetic.
- The voice can be synthetic. When no usable original recording exists, the Avatar speaks the person’s exact words in a synthetic stand-in voice — not a clone of theirs. We would rather say “synthetic” than let a warm sound imply something it isn’t.
- It only ever speaks the person’s words. When someone asks a question, the Avatar finds the passage in the person’s recorded stories that answers it and plays it back word for word. If a usable original recording of that passage exists, it plays the original. If not, it reads their exact words in that stand-in voice. It never composes a new sentence in their voice, never paraphrases, never summarizes.
- It shows its source. Every answer names the story it came from, so a listener can go and hear the whole thing.
- It admits what it doesn’t know. If the person never talked about something, the Avatar says so rather than guessing.
- Framing uses a different voice. A separate narrator — the Biographer voice the person chose for their sessions, never their own — can say a short, pre-written line to introduce or close an answer. That channel cannot carry story content.
- The person decides who can ask. Every story has its own audience — just the person, family, one specific person, or everyone — and the Avatar only answers from stories the listener is allowed to hear.
The honest limit is the same as the honest strength: the Avatar cannot say anything the person did not say. It will not offer new comfort, new advice, or an opinion on something that happened after they recorded. For many families, that is exactly the point. It is a way to ask a question and hear what they actually told you — not a guess at what they might say now.
Memorial Mode, which will let an archive be closed for good, is coming soon. It is meant for after the person is gone, and the person decides in advance who can do it: when it arrives, only a Guardian they chose will be able to request it. They will be emailed the moment a request is made, and there will be 72 hours to cancel it. The Avatar page describes the Avatar in more detail.
Questions worth asking before choosing any grief technology
Did the person say how they wanted to be remembered? This is the consent question, and it matters most. If they said they would not want to be simulated — or never had the chance to say — that belongs in the decision.
Does it say only their words, or new ones? Ask any service directly. If it generates replies, it is a simulation, however carefully it is made.
Who is this for, and what should it give them? Something that helps a person carry a loss is different from something that lets them avoid it, even when the screen looks the same.
What can still be captured while there is time? Recording someone’s actual voice and stories needs them to be here, and willing. For your own voice, the person who decides is you.
Recording someone while they are here
If you want to keep a real voice rather than build a model of one, start with consent and a single question. Ask first, make clear there is no pressure, and let some stories stay untold. Stop when they are tired; a short conversation that ends on their terms is still a good one.
How it works shows how a guided Quiddity session runs, and the pricing page shows which options include the Avatar. You can begin with a free five-minute warmup at quiddity.bio.