Grief produces an enormous, sudden amount of administrative and written work — a eulogy under deadline, a funeral program, dozens of thank-you notes, an obituary, sometimes legal and financial paperwork on top of all of it — at the exact moment a person has the least capacity to do it. AI tools can genuinely help with that specific, practical burden. They can also, with the same fluency, be prompted to generate messages written as if from the person who died, and that is a different thing entirely, one worth refusing regardless of how comforting it might feel in the moment.
This article draws that line as plainly as possible: what AI can help you write and organize during grief, and what it should never be asked to simulate.
What “griefbots” actually are, and why the line matters
A small but growing “digital afterlife” industry now offers services — sometimes called griefbots, deadbots, or AI ghosts — that train a chatbot on a deceased person’s texts, emails, voicemails, and social media to produce a simulated version of them you can converse with. A 2024 ethics paper in Philosophy & Technology mapped the risks across three groups: the “data donor” whose messages train the bot, the “data recipient” who controls it after death, and the “service interactant” who talks to it — and found that none of these relationships come with the kind of consent structure that would normally govern using someone’s likeness and words (Hollanek & Nowaczyk-Basińska, Philosophy & Technology, 2024). Their key recommendations: restrict these services to adult users, require actual consent from both the person whose data is used and the person interacting with the resulting bot, and build in a sensitive way to eventually “retire” the bot rather than leaving it running indefinitely.
Real-world cases show why the caution is warranted rather than theoretical. One widely reported case involved a man who built a chatbot of his deceased father from years of recorded interviews, later turning it into a commercial product; a psychologist consulted by the BBC about that story cautioned that “grief tech” used at a moment of heightened emotion needs to be approached slowly, because “you’d have to feel quite solid before using something like this” (BBC News, “The man who turned his dead father into a chatbot,” 2024). A 2025 Nature feature on the same industry quoted Nora Lindemann of Osnabrück University: with an ongoing simulated presence, “people don’t need to adjust to a world without this person. They can live in a somewhat pretend, in-between stage” (Nature, “Ready or not, the digital afterlife is here,” 2025). That is the mechanism this article asks you to avoid, regardless of which specific product or prompt produces it.
What honest AI use in grief looks like
The distinction that matters is not “any AI use during grief is wrong.” It is: AI can help you produce something that is yours — your memories, your words, your logistics — and it must never be asked to produce something that claims to be theirs.
Genuinely useful, low-risk uses:
- Organizing scattered notes and memories into a eulogy structure, using only what you actually wrote or said.
- Drafting a funeral program, an obituary from facts you provide, or a guest list and thank-you note tracker.
- Transcribing old voicemails, letters, or recordings for an archive — see archiving family photos and stories without inventing history for the consent and accuracy discipline that applies just as much to grief archives as to ordinary family ones.
- Helping you find words for a card or message to another grieving family member, when you supply the specific memory or sentiment and the model only helps you phrase it.
Never appropriate, regardless of how it is framed:
- Asking a model to write “what they would say to me” or generate a reply in their voice.
- Feeding a model someone’s messages or voice recordings to build an interactive persona of them.
- Any prompt structured as a conversation with the deceased person, even framed as “just for closure.”
- Using a companion app’s roleplay feature to simulate an ongoing relationship with someone who has died.
Do not use AI to generate séance-style prompts, an avatar of the person you lost, or any simulated reply “from” them, no matter how much comfort it seems to offer in the moment. There is no accountable consent structure for this — the person cannot agree to how their likeness is used after death, and no honest labeling of AI-generated text changes that. If you are drawn toward this specific use, that pull is itself a sign to talk to a grief counselor or trusted person rather than a chatbot. If you are having thoughts of suicide or self-harm, or you are worried about your immediate safety, contact your local emergency services now, or reach out for support through IASP’s crisis resources. Do not use a chatbot as a substitute for that step.
The safe-use table
| Task | Safe | Why |
|---|---|---|
| Organizing your own eulogy notes into a structure | Yes | The words and memories are yours; AI only organizes |
| Drafting an obituary from facts you supply | Yes | Factual, administrative, easily verified |
| Transcribing a voicemail or letter for an archive | Yes, with consent discipline | See the family archive consent framework for accuracy and permission norms |
| Reflection journaling about your own grief | Yes | You are processing your own experience, not simulating theirs |
| Requesting a message “as if from” the person who died | No | Impersonation with no consent structure |
| Building an interactive chatbot from their messages/voice | No | Digital afterlife industry’s own ethics research calls for adult-only, consent-based use with a retirement plan most consumer tools do not offer |
| Using AI to decide what to feel or how long grief “should” take | No | Not a clinical or therapeutic function AI can perform; see AI is not therapy |
The full grief AI safe-use table expands this with space to note your own specific tasks as they come up.
Why this is not a small caution
It would be easy to read the “never” column as overcautious, especially against a headline claim that some users report the interactions as comforting. That comfort is real for some people in the moment; it does not resolve the underlying problem. No available consent mechanism lets a deceased person agree to how their words and likeness get reused, no chatbot’s tuning makes it capable of continuing an actual relationship rather than generating plausible-sounding text, and the honest, evidence-grounded position from the researchers who study this space directly is that we do not yet know whether simulated ongoing “contact” helps grief or delays the adjustment grief requires. Given that uncertainty, the tasks in the “safe” column above are more than enough to get real, practical help from AI during an already overwhelming time, without introducing a use case nobody can yet vouch for.
Try it today
If you are facing a specific piece of grief-related writing this week — a eulogy, an obituary, a set of thank-you notes — use the safe-use table to structure how you ask AI to help: supply your own memories and facts, ask only for organization and phrasing help, and keep the final words recognizably yours. If any part of what you are tempted to ask for edges toward “make it sound like them,” stop and bring that specific instinct to a person — a grief counselor, a chaplain, or someone who knew them too — rather than to a chatbot.
