Jenny Kidd studies how societies remember. So when her research group set out to test the AI services that promise to let you keep talking to the dead, she did the obvious and slightly eerie thing: she fed one her own videos, messages and voice notes and built a griefbot of herself. The result did not feel like a ghost. It felt like a cheap impression, and that, she says, was the disturbing part.
“What was unsettling was seeing how a small amount of personal data could still produce something that superficially resembled a person, even while being clearly inaccurate,” Kidd, a media and cultural studies scholar at Cardiff University, told UNILAD Tech this week. Her verdict arrives as some of the best-known “digital immortality” companies quietly close their doors.
The short version
- Cardiff’s Jenny Kidd and King’s College London’s Eva Nieto McAvoy built “digital doubles” of themselves on four deathbot platforms for a peer-reviewed study.
- The bots copied verbal tics but lost track of ideas, contradicted themselves and met talk of death with cheerful emojis.
- The business is fragile: HereAfter AI posted a farewell notice in July 2026, and StoryFile went through Chapter 11 in 2024.
- Prices start at a few dollars, but the bigger costs are consent, data and who controls a simulation after you die.
- You can set your wishes now with a digital will, Apple’s Legacy Contact and Google’s Inactive Account Manager.
What the Cardiff and King’s researchers actually built
The work comes out of Synthetic Pasts, a project on how technology shapes personal and collective memory. In a paper published last October in the journal Memory, Mind & Media, Kidd and Nieto McAvoy ran what they call socio-technical walkthroughs of four services: Almaya, HereAfter, Séance AI and You, Only Virtual. They played both the person preparing a synthetic afterlife and the bereaved relative talking to one.
Some tools work like searchable archives of recorded stories. Others ingest messages, posts and voice samples and generate a chatbot that answers “in their tone and style.” In their write-up for The Conversation, the pair said the more they personalized the bots, “the more artificial it felt.” One bot, asked about the death it was meant to have experienced, replied:
“Oh hun… 😔 it (the death) is not something I’d wish for anyone to dwell on. It’s all a bit foggy now, to be honest. 🌫️ Let’s chat about something a bit cheerier, yeah?”
Kidd told UNILAD Tech the systems “were actually quite poor at maintaining coherence or internal consistency; they would contradict themselves, lose track of ideas, or generate responses that didn’t really resemble us.” What they could do was mimic “certain stylistic patterns or repeated lexical choices.” What came out, in her words, was “a kind of bland line of best fit,” and “something safe, smooth and, ultimately pretty weird.”
Her worry is less that a griefbot might be too convincing than that it reshapes the dead into something easier to sell. “There is a risk that conversational simulations flatten a person into a smoother, more compliant, and more platform-friendly version of themselves,” she said. “For some, that may be comforting. For others, it may be jarring or emotionally destabilizing.”
An industry built on grief, and on shaky ground
The idea has a well-known origin story. After her close friend Roman Mazurenko was killed in a car accident in Moscow in 2015, Eugenia Kuyda used thousands of their text messages to build a chatbot that talked like him. Strangers asked for bots of their own, and the project grew into Replika, launched in 2017.
In 2021 the San Francisco Chronicle told the story of Joshua Barbeau, who paid $5 for an account on Jason Rohrer’s Project December and fed it old messages from his late fiancée, Jessica Pereira. After the story went viral, OpenAI cut the service off from GPT-3 when Rohrer declined to add safeguards it requested. Project December is still online and still advertises access for $5.
Others took a more structured route. Former journalist James Vlahos built HereAfter AI out of the “Dadbot” he made of his dying father. Relatives could ask it questions and hear recorded answers in the person’s own voice. StoryFile filmed people answering hundreds of questions, a format used to preserve Holocaust survivors’ testimony. Justin Harrison founded You, Only Virtual after building what he calls a “Versona” of his mother, Melodi, before she died of cancer.
That timeline backs up one of Kidd’s sharper observations. “At the moment, many of these new platforms and services fail within a couple of years which would suggest the appetite to do death differently is not yet widespread,” she said. HereAfter, one of the platforms in her study, posted a notice on its homepage in July 2026 that read: “It’s been an honor to help families record their memories for loved ones, but unfortunately HereAfter is shutting down.” Users were told to email support for their recordings; by mid-August the site no longer loaded. StoryFile, which counted William Shatner among its backers, filed for Chapter 11 in May 2024. The case was dismissed in April 2025, and the company now pitches itself mainly to museums. If a parent’s recorded voice lives on a startup’s servers, the company’s survival decides whether the grandchildren ever hear it. Even big companion apps wobble: Character.AI filed for bankruptcy earlier this month.
What these services cost
The entry price is low. Before it closed, HereAfter charged $3.99 to $7.99 a month depending on how many stories you could record, with a 14-day free trial. Project December asks $5. In China, NPR found Silicon Intelligence selling a basic talking avatar of a dead relative for 199 yuan, about $30, built from less than a minute of good video and audio, while e-commerce listings offered to animate a still photo for under $2.
The money is the smaller question. In their Conversation piece, Kidd and Nieto McAvoy point out that these are “not memorial charities, they are tech start ups,” and that subscription fees, “freemium” tiers and partnerships with insurers or care providers “reveal how remembrance is being turned into a product.” A subscription can also outlast the person who bought it, as Cambridge researchers warned two years ago.
| Service | How it works | Last published price | Status, October 2026 |
|---|---|---|---|
| HereAfter AI | Voice-recorded life stories; relatives ask questions and hear answers | $3.99 to $7.99 a month | Shutting down; site offline since August |
| StoryFile | Pre-filmed video answers matched to spoken questions | Quoted per project | Exited Chapter 11 case in 2025; now focused on museums |
| Project December | Text chatbot that can simulate anyone, including the dead | $5 | Online |
| You, Only Virtual | “Versona” built from texts, voice memos and a relationship’s history | Not listed | Website shows a waitlist |
| Silicon Intelligence (China) | Talking video avatar from under a minute of footage | From 199 yuan (about $30) | Operating as of NPR’s 2024 report |
The Cambridge warning about “digital hauntings”
In May 2024, Tomasz Hollanek and Katarzyna Nowaczyk-Basińska of Cambridge’s Leverhulme Centre for the Future of Intelligence published “Griefbots, Deadbots, Postmortem Avatars” in Philosophy & Technology. In one hypothetical, a company called “Stay” lets an elderly parent sign up for a 20-year subscription before they die. Afterward, the bot emails the adult children in the parent’s voice. One child refuses to engage; the other ends up emotionally exhausted and guilty about what will happen to the bot, and neither can switch it off because the contract forbids it.
“These services run the risk of causing huge distress to people if they are subjected to unwanted digital hauntings from alarmingly accurate AI recreations of those they have lost,” Hollanek said at the time. “The potential psychological effect, particularly at an already difficult time, could be devastating.” The pair called for consent from the person recreated, clear AI disclosure, and ways to “retire” a deadbot, something like a digital funeral. Nowaczyk-Basińska put it bluntly: “We need to start thinking now about how we mitigate the social and psychological risks of digital immortality, because the technology is already here.”
Who owns the data of the dead?
Nobody has fully settled it. Europe’s GDPR explicitly does not apply to the personal data of deceased people (Recital 27), though it lets each country write its own rules. France did in 2016, letting people register directives on how their data should be kept, deleted or shared after death. Most US states have adopted the Revised Uniform Fiduciary Access to Digital Assets Act, which covers executors’ access to a dead person’s accounts. None of those laws was written with an AI replica in mind. In China, NPR found that Super Brain and Silicon Intelligence require authorization from the person or proof of kinship, but that is company policy, not law.
Kidd expects the bigger shift to happen outside dedicated grief startups. “For now, it is probably more likely that we will see people use their preferred conversational LLM system as a mechanism to do this work,” she said, naming ChatGPT, Claude and Meta AI. If you go that route, it is worth knowing what a chatbot company keeps and who can read it before you paste in years of a loved one’s messages.
What you can do before anyone builds one
- Decide whether you want a griefbot at all, and write it down. A line in your will or a letter to your executor gives your family something to point to.
- Set up Apple’s Legacy Contact. On an iPhone running iOS 15.2 or later, go to Settings, tap your name, then Sign-In & Security, then Legacy Contact. Your contact will need the access key and a death certificate, and gets three years of access from the first approved request. Purchased media and Keychain passwords are excluded.
- Set up Google’s Inactive Account Manager. You choose how long Google waits after you stop using your account, name up to 10 trusted contacts, and pick which data each one can download. Tell them to expect a time-limited download link.
- Keep your own copies. If you record stories with a startup, download the files regularly.
- Before building a bot of someone else, ask who else it affects. Siblings, partners and children may not want to hear that voice again. Check that you can pause, export and delete the persona.
Photographs and home movies have always done some of this work, and Kidd notes AI “supercharges this landscape.” Her point is narrower and more useful: the chatbots on offer today do not hold a person, they hold a pattern. Whether that pattern brings comfort or reopens a wound depends on the family, and the people deciding should be the ones grieving, not a subscription screen.
Sources and further reading
- UNILAD Tech: Researcher who ‘immortalized’ herself in an AI bot reveals most ‘unsettling’ thing she found
- The Conversation: Can you really talk to the dead using AI? We tried out ‘deathbots’ so you don’t have to
- Cardiff University ORCA: Synthetic afterlives: deathbots as affective infrastructures of memory (Memory, Mind & Media)
- The Debrief: Cambridge researchers warn of digital ghosts of dead loved ones
- NPR: Chinese companies offer to ‘resurrect’ deceased loved ones with AI avatars
- Apple Support: How to add a Legacy Contact for your Apple Account
- Google Account Help: About Inactive Account Manager
About this article: GeekBlog covers U.S. technology news, AI, phones, smartwatches and gaming. Every story is written and checked under our Editorial Policy. Spotted a mistake or have a story tip? Contact our editors.
