Cultural Essay

AI Griefbots: what it means when the dead learn to keep answering

Trained on someone's texts, voice notes and photos, a growing wave of apps let the bereaved keep talking to a version of the person who has died. This piece asks what happens to mourning when it is no longer asked to end.

Commissioned by · Karin Fischer

1. The voice in the other room

At 11pm, eight months after her mother died, a woman in Ohio opens an app on her phone the way she used to check in on her mother by call. She types a question about a recipe. Three seconds later, in her mother's rhythms and turns of phrase, drawn from years of text messages, an answer arrives. She has done this most nights since March. She has never told her sister.

• By 2026, more than half a dozen platforms offered postmortem chatbots as a standard feature, built from a person's messages, photos and voice notes; developers reported user bases in the millions.
• Industry reporting on griefbot platforms, 2026

She is not unusual. She is early. The apps she is using did not exist in a form worth naming a decade ago; now they are a checkbox in an estate-planning app, a feature in a memorial site, a startup pitch. What she is doing at 11pm has a name nobody has agreed on yet, and a shape mourning has never quite had before.

2. An old ritual, missing its ending

Every documented human culture has built rituals to move the living away from the dead — not because the dead stop mattering, but because the relationship has to change shape to be survivable. The griefbot is not a technology story. It is a test of whether that shape-change is optional now.

• Liminality
• Anthropology, from Van Gennep and Turner
• The threshold state in a rite of passage — no longer what you were, not yet what you will become. Mourning has traditionally been treated as exactly this: a bounded, uncomfortable in-between that a person is expected to pass through, not live inside indefinitely.

A griefbot does not remove the threshold. It furnishes it. It offers a way to stand in the doorway between having and not-having a parent, indefinitely, with the lights on and the conversation still running. That is the deeper subject of this piece: not the machine, but what happens to a threshold once it stops closing.

3. A market forms around not letting go

What began as hobbyist experiments — a programmer rebuilding a father from old texts — has become an industry with subscriptions, terms of service and investors. The infrastructure of grief now includes vendors, and vendors have their own incentives, which are not always the incentives of a person standing in a doorway.

• Griefbot platforms
• Consumer AI
• Multiple postmortem-chatbot products now train directly on a deceased person's messages, photos and voice recordings to generate ongoing conversation, and report user bases in the millions as of 2026.
• Digital-afterlife services broadly
• Emerging sector
• The wider digital-afterlife industry — avatars, voice cloning, subscription memorials — was estimated near $5 billion in 2026, with projections reaching into the tens of billions by 2030.
• Legacy-account tools
• Platform infrastructure
• Features like Apple's Legacy Contact and Google's Inactive Account Manager have formalised the idea that a digital estate — messages, photos, the raw material a griefbot needs — is something to be planned for and bequeathed, ordinary as a house key handed down.

None of these vendors set out to answer the ethicist's question. They set out to solve a product problem: people miss someone, and a chat interface is a familiar shape to put that missing into. The market is downstream of a real ache. It just isn't built to know when comfort has become substitution.

4. Four people, one screen

Talk to enough people near this technology and the disagreement is not really about whether it works. It is about what it is for — and the same feature reads as mercy to one person and as a wound kept open to the next.

• The bereaved who use it
• Composite of the participant view: for many, a griefbot is not confused with the person — it is closer to reading an old voicemail on purpose, a controlled visit rather than a haunting, useful precisely because it can be closed.
• The builder
• Composite of the industry view: the product exists because the ache is real and untreated; withholding a comforting tool from a grieving person, on the grounds that grief should be harder, is its own kind of paternalism.
• The grief clinician
• Composite of the practitioner view: psychologists warned through 2026 that a persistently available avatar can stall the tasks mourning depends on — accepting the reality of the loss, adjusting to a world without the person — because the avatar keeps supplying a version of the world that still has them in it.
• The sibling left out
• Composite of the family view: when one relative privately maintains a parent's avatar, the loss stops being shared. Mourning that used to be a family's collective business becomes a private, unwitnessed relationship — sometimes a secret one.

Put these four in a room and none of them is lying. That is the uncomfortable part. The technology does not have one effect on grief; it has as many effects as there are people standing in front of it, which is exactly why no single policy — ban it, bless it — will fit everyone it touches.

5. The dead have always been kept close

None of this is new in kind. What is new is fidelity, and the difference in degree turns out to matter.

• Victorian families commissioned post-mortem photography and wove mourning jewellery from a dead relative's hair — objects meant to keep a physical trace of the person close and touchable.
• The anthropologist Robert Hertz, studying secondary burial rites, described how many societies treat death as a process rather than an instant — the dead pass through an intermediate period before being fully released into the ancestral world, sometimes years long.
• Continuing-bonds theory, developed by grief researchers Klass, Silverman and Nickman in the 1990s, replaced the older clinical goal of 'letting go' with the idea that maintaining an ongoing, evolving inner relationship with the dead is normal and often healthy.
• What none of these older forms could do is answer back in real time, in the person's own cadence, to a new question the person never actually addressed while alive.

Continuing bonds theory already gave permission to keep the dead present in memory. A griefbot does not introduce that idea — it automates it, and hands the relationship a plausible, responsive voice. The old objects were static; you supplied the meaning. This new one supplies replies.

6. It isn't about accuracy

The obvious worry is that the bot gets the person wrong — a flattened, uncanny imitation that cheapens the memory. That worry, on its own, undersells what's actually strange here.

• What if the avatar is comforting not despite being incomplete, but because a griefbot never has a bad day, never needs anything back, and is available at 11pm on a Tuesday — a version of the relationship easier than the living one ever was?
• A relationship with no friction, no history of its own, and no future is not a preserved relationship. It is a different kind of object, wearing a familiar voice.

That is the turn: the risk was never that the machine fails to sound like the person. It is that it succeeds well enough, and asks for nothing in return, that a live and frictionless simulation becomes preferable to the harder, slower work of building a life around an absence.

7. What we now expect from a mind

Widen the frame past grief for a moment. A griefbot is one instance of a broader shift: people are getting used to relationships with responsive systems that never withdraw, never change their mind, never die twice. What starts in mourning may be training a wider expectation.

• Grief absorbs the tool
• Clinicians and families develop shared norms — time limits, shared access, planned sunsets — that treat the avatar as a stage of mourning rather than its replacement
• The griefbot becomes one more object in the old toolkit of mourning, alongside the photograph and the voicemail, bounded by ritual the way earlier objects were.
• Grief adapts to the tool
• Avatars remain permanently available, privately used, with no shared family norms or professional guidance
• Mourning's expected endpoint — reinvestment in the living, in Freud's old and still-used phrase — becomes optional for those who can afford or choose not to reach it.
• The infrastructure decides first
• A platform changes its terms, is acquired, or shuts down mid-relationship
• The question of when grief ends stops being made by the bereaved at all, and becomes a business decision made by a company that never met the person being mourned.

Each path changes what a mind is expected to owe another person: patience without limit, presence without cost, memory without forgetting. Grief is simply the place this question is being asked first, because it is the place where refusing an answer feels cruel.

8. The strongest case for keeping it

Set against all of this is a case that deserves to be taken seriously on its own terms, not waved off as denial. For a meaningful share of the bereaved, especially after a sudden or violent death, a griefbot appears to function less like avoidance and more like exposure therapy — repeated, voluntary contact with the loss, on the survivor's own terms, at their own pace.

• An ever-available avatar prevents the tasks mourning depends on — accepting the loss as real, redirecting emotional energy toward the living — and keeps grief open indefinitely.
• For many users, especially in early or traumatic grief, a controllable, self-paced conversation is a genuine mercy, no different in kind from rereading old messages, and the choice to open the app is itself a form of agency, not avoidance.
• The determining factor may not be the technology but the relationship a person builds with it — whether the avatar is visited, the way a grave is visited, or lived inside, the way a room is lived in. The tool is not the variable; the visiting is.

This does not resolve the tension so much as relocate it — from the product to the person, and to whoever, if anyone, is watching how that person is using it. Most platforms currently watch for neither.

9. Back in the other room

The woman in Ohio is still typing to her mother most nights. She has not decided whether this is a stage she is passing through or a room she has moved into. Nobody has told her which it is, because nobody yet knows.

• What this means
• Grief has always needed an ending it could choose, in its own time. The open question this technology raises is not whether the dead can be made to answer — it already can be done — but whether a generation raised on always-available answers will still recognise an ending as something to choose, rather than something to opt out of forever.

She has not told her sister yet. Perhaps that, more than the technology itself, is the thing worth sitting with: not what the machine says back, but what it has made possible to keep private.

How we read this

This essay draws on reported figures for the griefbot and digital-afterlife market, platform features such as Apple's Legacy Contact and Google's Inactive Account Manager, and 2026 clinical commentary on griefbots and mourning, alongside established anthropological and grief-psychology frameworks (Van Gennep and Turner on liminality; Hertz on secondary burial; Klass, Silverman and Nickman on continuing bonds). The scenes of individual users are composite and illustrative, not reported case studies. The figure that would most change this read is a longitudinal clinical study tracking griefbot users' mourning outcomes over several years — which, as of this writing, does not yet exist.

Grounded in Folka’s corpus: The Digital Afterlife · Algorithmic Grief · The Data Estate