Talking to the Dead: Inside the AI 'Griefbots' Industry, and What to Know Before You Try One

Feed an AI a person's old texts, voicemails, emails, and a handful of photos, and it can hold a conversation in their voice - respond to "how was your day" the way they used to, remember the nickname they had for you, even show up on a video call and nod along. The person on the other end of that conversation has been dead for months, sometimes years. This isn't a thought experiment anymore. More than half a dozen companies now sell exactly this service, and researchers say millions of people have already used one to talk to a recreation of someone they lost.

That's a genuinely strange sentence to write, and it's also just where a chunk of the AI industry is right now. The tools are getting good enough that a real academic field - grief tech, or the "digital afterlife industry" - has formed around the ethics of it, and the research on whether it helps or harms is still actively being written.

How it actually works

The pattern across platforms like HereAfter AI, Seance AI, You Only Virtual, and Re;memory is the same one behind most modern AI products: feed a model enough of a person's real material, and it can generate plausible new material in that person's style. Here, the training data is deeply personal - old text message threads, voicemails, social media posts, emails, home videos, sometimes a structured interview the person recorded about their own life before they died. The output ranges from a simple text chatbot to, on the higher end, an animated avatar that can smile, turn its head, and speak in a voice cloned from a few minutes of audio.

Researcher Jed Brubaker at the University of Colorado Boulder, who studies this space, ran a 2026 study and found the bar for a convincing result is lower than most people expect: participants filled out a short survey sharing a few memories and personality traits about someone they'd lost, and a single prompt to a free ChatGPT account produced what Brubaker's team calls a "generative ghost" - a thin but recognizable representation, built from almost nothing. The dedicated grief-tech platforms go much further with dedicated training data, but the underlying finding still matters: you don't need a specialized company or a large budget to create one of these. Anyone with a phone full of old messages already has the raw material.

The market reflects how fast this has moved from novelty to business. One estimate puts the digital-afterlife industry at roughly $22 billion in 2024, projected to reach $80 billion by 2034. Pricing is already consumer-subscription simple: Re;memory charges around $24 a month for up to three custom avatars, and Seance AI starts near $20 a month for an animated version of a loved one, on top of a free tier that offers a text-only "seance."

Why people are actually using it

The use cases split less neatly than you'd think - this isn't just grief-stricken people refusing to let go.

Processing loss with something that talks back. A June 2026 study published at the Designing Interactive Systems Conference found that some participants genuinely valued the chance to have one more conversation - to ask a parent for advice they wished they'd gotten in time, or to say something left unsaid. For a specific kind of unfinished business, a chatbot that responds in a familiar voice can create a sense of closure that journaling or therapy alone doesn't replicate.

Preserving a life story before it's gone. HereAfter AI's core pitch isn't actually post-mortem - it's aimed at people while they're still alive, recording a structured "Life Story Avatar" through guided interview questions so that grandchildren can one day ask a grandparent's AI what their own childhood was like. Framed this way, it's closer to an interactive oral history than a resurrection.

A single low-stakes moment, not an ongoing relationship. Plenty of users described in coverage of this space are trying it once - a birthday, an anniversary of a death, a specific unresolved conversation - rather than building a standing relationship with the recreation. That distinction turns out to matter a lot for how researchers think about risk, which is the next part of this story.

Where it gets complicated

This is the part worth slowing down for, because the concerns aren't hypothetical anymore - they're showing up in actual bereavement care and actual research.

It can complicate grief instead of resolving it. Hospice industry coverage in April 2026 flagged that griefbots reflect a user's own emotions back at them, which can feel validating in the moment but may reinforce a person's inability to accept a loss rather than help them move through it - particularly for someone already in a vulnerable state. A separate line of research on "deadbots" found the same pattern: continuous, on-demand access to a simulated version of the deceased can blur the line between memory and ongoing presence in ways that make some people's grief more pathological, not less.

Consent is mostly unresolved. Most scholars in this field, and 95% of respondents in one survey, oppose creating a griefbot of someone who never agreed to it while alive. As of 2026, there's no consistent legal framework preventing it anyway - nothing stops a grieving family member, or an ex-partner, or a stranger with access to someone's public social media, from building a chatbot version of a dead person who never consented to being recreated.

Families don't always agree with each other. Hospice News reported a recurring conflict: one family member finds real comfort in a bot recreation of a parent, while a sibling finds the same bot disturbing or disrespectful. There's often no single "right" answer for a family, and no established norm yet for whose preference wins.

It's not the same category as a deepfake, but it borrows the same techniques. Brubaker's research draws a real distinction here - a deepfake's defining feature is intent to deceive, and most people using a griefbot know exactly what it is. But the underlying capability (a cloned voice, a generated likeness, a model trained on someone's real communications) is identical to what makes deepfakes possible, and a platform with weak safeguards could be pointed at either use.

The emotionally vulnerable are a predictable target for upselling. Multiple researchers have flagged the same commercial risk we've written about with AI companion chatbots: a person in acute grief is not in a great position to evaluate whether a $20-a-month subscription is a healthy coping tool or something they'll regret continuing to pay for a year from now.

What to think through before you try one, or make one

1. Decide if this is a one-time moment or an ongoing habit, and be honest with yourself about which one you're building. The research suggests a single meaningful conversation and a standing daily relationship carry very different risks - know which one you're signing up for.
2. If the person is still alive, ask them. HereAfter AI's model - recording your own life story for your own future avatar - sidesteps the consent problem entirely. If you're building a version of someone who's already gone, there's no way to get their consent after the fact, which is exactly the gap most of the ethical concern is about.
3. Check with the rest of the family first, not after. Given how often this creates conflict, treat it the way you'd treat any other decision about a shared loved one's memory or estate - as something to discuss, not something to surprise people with.
4. Look at what happens to the data. These platforms are built on some of the most sensitive material that exists - private messages, a cloned voice, a person's likeness. Before uploading it, check the company's data policy the way our AI Safety & Privacy Checklist recommends for any tool handling personal information, and ask what happens to that data if the company shuts down or is acquired.
5. If grief feels like it's getting harder to sit with rather than easier, that's worth noticing. A griefbot that leaves you feeling more able to accept a loss is doing something different than one that leaves you dreading the day you can't talk to it anymore. If it's the second one, a real conversation with a grief counselor or therapist is worth having alongside it, not instead of it.

Key takeaways

• AI griefbots are a real, growing industry, not a fringe experiment - a market estimated at $22 billion in 2024 and projected to reach $80 billion by 2034, built on more than half a dozen consumer platforms.
• The technology bar is lower than people assume - a single prompt with a few shared memories can produce a recognizable "generative ghost," which means anyone with old messages and photos already has what they'd need.
• The research is genuinely mixed, not settled - some users report real comfort and closure, while hospice and grief researchers warn the same interaction can reinforce unresolved grief for others.
• Consent and family agreement are the two biggest unresolved problems - most experts oppose recreating someone without their consent, there's no law consistently preventing it, and families frequently disagree on whether a recreation should exist at all.
• This sits in the same ethical territory as other emotionally sensitive AI tools - the same questions about data, vulnerability, and disclosure that apply to companion chatbots apply here, just with higher emotional stakes.

If you're building general AI literacy before going deeper on any one tool, AI 101 is the place to start. For the broader landscape of AI in emotional and mental health contexts, Mental Health covers related tools and considerations. And before trusting any platform with someone's private messages, voice, or likeness, our AI Safety & Privacy Checklist is worth reading first.

info

This content was developed with AI assistance and is regularly reviewed for accuracy. Market projections, study findings, and platform pricing reflect reports published through September 2026 and may change as this industry and its research base continue to develop.