Digital LegacyBy Stephanie Werner · August 2026 · 6 min read

The Dead Are Texting Back: Inside the Grief Bot Industry

AI companies will now build a version of your dead person you can talk to. Some grievers call it comfort. Researchers are calling for guardrails. Nobody knows yet what it does to grief.

A woman lit by her phone screen at night beside a framed portrait of a person, in black and white

Last verified July 30, 2026.

One night in the fall of 2020, a freelance writer named Joshua Barbeau typed a sentence into a chatbot and his fiancée answered. She had been dead for eight years [1].

He had fed a language model her old texts and a short description, using a service called Project December. The conversations that followed, reported at length by the San Francisco Chronicle, read the way ghost stories have for centuries: intimate, uncanny, comforting, wrong, right. He knew it wasn’t her. He kept typing anyway.

Since then, that experiment has become an industry. This is the tour: who’s selling what, what psychology says, what researchers are warning, and the one thing you should decide on purpose, in writing, while you’re alive.

The product shelf

The offerings sort into two very different species, and the difference is consent.

The first species records the living. HereAfter AI interviews you, over hours, about your life, then packages the answers into an interactive “legacy avatar” your descendants can question [2]. StoryFile did the same with video, its technology famously letting Ed Asner field questions at his own memorial service [2]. You participate. You curate. You know.

The second species reconstructs the dead. Feed a model the texts, voicemails, emails, and posts a person left behind, and it generates them: their phrasing, their jokes, their typos. Project December sits here. So did the origin story of Replika, which began when its founder rebuilt her closest friend, Roman Mazurenko, from his messages after he was killed in an accident [3].

In China, a full commercial sector now sells “digital resurrection,” grieving parents commissioning video avatars of dead children for modest fees, and thousands of people have already paid for them [4] [5].

The dead person, notably, signed off on none of species two.

Why it works, and why that's the question

Grief science has a concept called continuing bonds: the finding, mainstream since the 1990s, that healthy grieving usually isn’t about letting go [6]. People keep talking to their dead, at gravesides, in journals, in their heads, for decades, and mostly it helps. Seen through that lens, a grief bot is just a more responsive version of the same thing, and some users describe exactly that: a place to say the unsaid things, a bridge through the worst months.

But continuing bonds research came from bonds that couldn’t answer. A bot answers. It’s available at 3 a.m., it never tires of you, and it’s engineered, like all engagement products, to keep the conversation going. Clinicians’ concern has a name: prolonged grief disorder, added to psychiatry’s diagnostic manual in 2022, the roughly one-in-ten cases where mourning locks up instead of moving [7] [8].

Does a responsive simulation of the dead help the stuck get unstuck, or does it keep them stuck? Here’s the honest answer: nobody knows. The long-term studies don’t exist yet. Every user is the experiment.

The warnings on the record

The most cited academic work on this so far came out of Cambridge in 2024: two researchers at the Leverhulme Centre for the Future of Intelligence published an analysis of the “digital afterlife industry” in the journal Philosophy & Technology, and their scenarios are worth reading before you upload anything [9].

They warn about consent, in both directions: the dead who never agreed to be simulated, and the living, including children, who may be confused by a dead parent’s avatar, as in their scenario of a terminally ill mother who leaves a chatbot for her eight-year-old son [10]. They warn about what they call unwanted hauntings: mourners who want to stop and find that the service, or a contract the dead person signed, won’t let them [10].

They flag the obvious commercial nightmare: an avatar of your grandmother that starts recommending products once a free trial ends [10], because engagement is the business model and grief is a very sticky user state. And they call for guardrails the industry mostly lacks: age restrictions, transparency, and dignified retirement protocols, ways for a bot to be laid to rest [9].

StoryFile, the company that animated Ed Asner, filed for bankruptcy in 2024 [11]. That raises the question every one of these products carries in its terms of service: your dead person’s avatar isn’t an heirloom. It’s a hosted service. When the company dies, so, again, does the person, and the family grieves the loss again, this time because a company shut down.

The clause nobody's written yet

Here’s where this lands in MORTL territory: the raw material of a grief bot is your data, your texts and voice and face, and right now, in most places, the decision to animate it after your death belongs to whoever holds the archive. Which means it’s an estate planning question that almost no estate plan answers.

Before uploading anything, ask who owns the source files and generated model, whether the company uses them to train other systems, how consent is documented, who can delete the data and simulation, what happens after cancellation or bankruptcy, whether the product can impersonate the person outside the memorial context, how minors are protected, and what mental-health or crisis safeguards exist. Save the current terms and privacy policy with the decision.

In your digital legacy documents, record whether you consent to a memorial simulation, which data may be used, who controls it, who may see it, and when it must be deleted. Enforceability varies by jurisdiction and contract, so a lawyer should review language intended to create legal duties.

If you’re considering using a grief bot, remember that responses are generated by a model and are not evidence of the deceased person’s current thoughts or presence. Stop using it and seek human support if it increases distress, isolation, compulsive use, confusion, or thoughts of self-harm. Grief doesn’t follow a fixed sequence, and people can respond to this technology differently.

The bottom line

The technology to make the dead text back exists, it’s for sale, and the research on what it does to the living is years behind the product. Two decisions are yours to make now: whether your own data can ever be used this way, in writing, and whether a simulation belongs anywhere in your grief. Both deserve to be choices. Right now, for almost everyone, they’re defaults.


Sources: San Francisco Chronicle, “The Jessica Simulation” (2021); Hollanek & Nowaczyk-Basińska, “Griefbots, Deadbots, Postmortem Avatars,” Philosophy & Technology (2024); DSM-5-TR, prolonged grief disorder (2022); Klass, Silverman & Nickman, Continuing Bonds (1996); reporting on StoryFile’s Chapter 11 filing (2024) and China’s digital resurrection industry (2023-2024). See our Sources & Methodology.

This article is education. If you're struggling, a grief counselor is a human who can actually hear you. Find support. If you're having thoughts of self-harm, call or text 988, the Suicide and Crisis Lifeline, any hour; reaching out is a reasonable step.

Sources & References

Research & Citations

All factual claims in this article are sourced from peer-reviewed research, government data, and named institutions. Citations follow APA 7th edition format.

  1. [1]Fagone, J. (2021, July 23). The Jessica simulation: Love and loss in the age of A.I. San Francisco Chronicle. ↗ Source Retrieved July 28, 2026
  2. [2]Carballo, R. (2023, December 18). Using artificial intelligence to talk to the dead. The New York Times (republished by The Seattle Times). ↗ Source Retrieved July 28, 2026
  3. [3]Newton, C. (2016, October 6). Speak, memory: When her best friend died, she rebuilt him using artificial intelligence. The Verge. ↗ Source Retrieved July 28, 2026
  4. [4]Yang, Z. (2024, May 7). Deepfakes of your dead loved ones are a booming Chinese business. MIT Technology Review. ↗ Source Retrieved July 28, 2026
  5. [5]South China Morning Post. (2024, February 10). Mourning relatives in China pay for AI avatars to 'resurrect' dead loved ones. ↗ Source Retrieved July 28, 2026
  6. [6]Klass, D., Silverman, P. R., & Nickman, S. L. (Eds.). (1996). Continuing bonds: New understandings of grief. Routledge. ↗ Source Retrieved July 28, 2026
  7. [7]American Psychiatric Association. (2022). Prolonged grief disorder (DSM-5-TR fact sheet). ↗ Source Retrieved July 28, 2026
  8. [8]Lundorff, M., Holmgren, H., Zachariae, R., Farver-Vestergaard, I., & O'Connor, M. (2017). Prevalence of prolonged grief disorder in adult bereavement: A systematic review and meta-analysis. Journal of Affective Disorders, 212, 138–149. ↗ Source Retrieved July 28, 2026
  9. [9]Hollanek, T., & Nowaczyk-Basińska, K. (2024). Griefbots, deadbots, postmortem avatars: On responsible applications of generative AI in the digital afterlife industry. Philosophy & Technology, 37, Article 63. ↗ Source Retrieved July 28, 2026
  10. [10]University of Cambridge. (2024, May 9). Call for safeguards to prevent unwanted 'hauntings' by AI chatbots of dead loved ones. ↗ Source Retrieved July 28, 2026
  11. [11]AI Business. (2024, May 8). Startup behind AI William Shatner files for bankruptcy. ↗ Source Retrieved July 28, 2026

Share this story

Post it, or send it straight to the person who needs it. The link goes with it.

Keep Reading

The Death Dispatch

More stories like this, weekly.

A newsletter about death you'll actually look forward to reading.