What is a griefbot, and can AI bring back the dead?

Griefbot: an AI chatbot, voice or avatar built to imitate a person who has died, usually from their messages, posts, photos and voice recordings, and usually by someone else after the death.

Deadbot and deathbot: other names for a griefbot, most common in academic writing.

Generative ghost: researchers' term for an AI of a dead person that generates new content instead of replaying what the person actually said.

Persona: the AI version of you that you build on Afterlife.ai™ while you are alive, from your own memories, values and cloned voice.

Executor Lock™: Idy's patent-pending system of named executors, Trusted Contacts, a verified death report and a sealed snapshot of your Persona that can never be changed.

Digital executor: the person you name to carry out your wishes for your accounts, files and AI likeness after you die.

A griefbot is an AI chatbot built to simulate a deceased person, so that the people left behind can type or talk to an imitation of them. Most griefbots are trained or prompted on the material a person leaves behind, such as text messages, social media posts, emails, voice recordings, photos and videos. Some answer in text, some speak in a cloned voice, and some appear as a talking video avatar or a virtual-reality figure. The idea entered wider public awareness through Black Mirror's 2013 episode "Be Right Back", in which a grieving woman pays for an AI version of her late partner, and it became a real product category within a decade.

The three common names describe the same thing. "Griefbot" is the word most journalists and most searchers use, "deadbot" and "deathbot" are more common in research papers, and Wikipedia files all three under one entry (Wikipedia, Deadbot). Researchers Meredith Ringel Morris and Jed Brubaker added a fourth term in 2024, "generative ghosts", for agents "capable of generating novel content rather than merely parroting content produced by their creator while living" (arXiv, January 14, 2024). That distinction between new words and the person's own words is the single most important idea on this page, because a griefbot that generates new sentences is always speaking for someone, and the question is who gave permission.

The scale of the raw material is enormous. Researchers Carl Öhman and David Watson at the Oxford Internet Institute projected that at least 1.4 billion Facebook users will die before 2100, and that the number could pass 4.9 billion if the network keeps growing (Big Data & Society, 2019). Every one of those accounts is a possible training set for a griefbot that its owner never approved. Meta has already patented a way to simulate dead users, with a model trained on a person's posts, likes and messages that keeps acting as that person after death. For a shorter definition page, read what a deadbot is.

No. AI cannot bring back a dead person, their mind or their judgment, and no honest company claims otherwise. What AI can do is reproduce patterns from what a person left behind, such as their phrasing, their favorite stories and the sound of their voice, and then generate new sentences that resemble them. The quality of the result depends almost entirely on the quality of the material and on whether the person chose that material while alive. A griefbot built from a few hundred social media posts produces a thin impression, while a Persona built from hours of the person's own guided storytelling carries their real memories in their own words.

People have wanted this for a long time. Scientific American, which ran a 1920 interview in which Thomas Edison discussed an apparatus for contacting the dead, reported in November 2025 that griefbot companies now operate across Asia, Europe and North America (Scientific American, November 18, 2025). The difference in 2026 is that the tools work well enough to fool the ear and the eye. That is why the Arizona appeals court drew its line where it did, between "the family's belief of what the victim would have said" and the person's "own voice and opinions".

New research shows what happens when that line blurs. In 26 in-depth interviews with people who use deadbots, Yifan Li and Xingyu Lan found that users shape the bot through selective input and ongoing adjustment, and that over time "AI-generated new memories blur with authentic recollections" (arXiv, March 1, 2026). A family that builds a griefbot after a death fills the gaps with its own hopes, and the bot hands those hopes back as if the dead person had said them. A Persona avoids that problem by design, because every memory in a Persona comes from the person, recorded while the person was alive and able to choose what to share.

Griefbots are made in two ways, by families using general AI tools and by companies selling the service. In the do-it-yourself version, a relative uploads messages or a written profile of the dead person into a general chatbot and asks the chatbot to answer as that person, and a free account is enough to produce something convincing. In the paid version, a company collects photos, videos and voice notes from the family and builds a talking avatar, a voice line or a chatbot. Both versions share one feature, which is that someone other than the dead person decides what the imitation says. The raw material increasingly includes a person's AI chat history, which no major assistant lets anyone inherit, as what happens to your ChatGPT memories when you die explains.

The history of the category is short and well documented. Eugenia Kuyda built a chatbot from the messages of her late friend Roman Mazurenko in 2016, and that project grew into the companion app Replika. James Vlahos recorded 91,970 words of his dying father's life story and turned them into the "Dadbot", which Wired published on July 18, 2017 and which became HereAfter AI (Wired). Ray Kurzweil built a "Fredbot" of his father in 2018, and the game designer Jason Rohrer launched Project December in 2020, which let people talk to language-model versions of loved ones (Scientific American). Researchers built their own versions too, including Muhammad Aurangzeb Ahmad's "Grandpa Bot" at the University of Washington and Hossein Rahnama's Augmented Eternity work at the MIT Media Lab (Wikipedia, Deadbot).

The largest technology companies have planned for this as well. Microsoft holds a US patent, granted December 1, 2020, on "creating a conversational chat bot of a specific person" (Google Patents, US10853717B2). Meta was granted a patent in late 2025 on simulating a user of a social network with a language model, including a user who is away or has died, and told Business Insider it had no plans to move forward with it (MediaPost, February 17, 2026). When the platforms that hold your messages own patents on ways to run you after death, the open question is whose rules apply.

Griefbot companies fall into three groups, and most of them build from what the dead left behind after the death. The first group sells talking avatars or voice lines to families, such as China's Silicon Intelligence and Super Brain, which MIT Technology Review reported had about 1,000 clients each by 2024, with tablet avatars at Super Brain costing $700 to $1,400 (MIT Technology Review). The second group offers chatbots that a user configures, such as Project December and Seance AI, both named in Scientific American's November 2025 report. The third group recorded people while they were alive, such as HereAfter AI and StoryFile, and both of those show how fragile a company can be.

HereAfter AI posted a shutdown notice in July 2026 and its website has been offline since August 2026, so families now have to email the company to ask for recordings (what happened when HereAfter AI shut down). StoryFile filed for Chapter 11 in May 2024 and was sold in March 2025. Eternos, the company that helped Michael Bommer build an AI of himself, renamed itself Uare.ai in November 2025 and moved into professional AI (TechCrunch, November 11, 2025). A Lehigh University audit of 93 AI "afterlife" services found that 82 publish thin or no information on who has authority over a recreation (arXiv, August 28, 2026).

Afterlife.ai™ belongs in none of those groups. Every Afterlife.ai™ Persona is built by the person the Persona represents, while that person is alive, and our Terms bar anyone from building a Persona of another person without that person's explicit written consent (section 5.1 of our Terms of Service). The person names the people who take over, the death is verified before anything changes, and the Persona is sealed exactly as the person left their Persona. For a comparison of AI twin and clone platforms, read the best digital twin apps compared.

Griefbots are already in courtrooms, on national television and in family living rooms across at least six countries. The cases below are the ones most often cited by journalists and researchers, and each one teaches the same lesson about consent and authority. The final column matters most, since who made the decision predicts how each story ended.

Two of these cases show the problem most clearly. Qiao Kangqiang described the unauthorized videos of his son as "rubbing salt in his wound", and in the United States Zelda Williams asked people in 2025 to stop sending her AI videos of her father, Robin Williams, then on September 24, 2026, after a fake clip spread online, asked again for people to leave him alone (Wikipedia; NZ City). Robin Williams had restricted the commercial use of his likeness until 2039 through his trust, and strangers made fakes anyway. A ban stops licensed uses, and it leaves a family with no authorized version of the person to point to.

Michael Bommer is the counterexample, and he is the model every family can follow. He made the decision himself, while he was alive, and he spent about 25 hours over two months telling his own stories in his own words. The company that built his AI later changed direction, which is why a person's decision needs governance written into the contract and enforced by the product itself, so the decision survives whatever happens to the company hosting the recording.

A griefbot is ethical when the person it imitates consented while alive and named who controls the recreation afterward, and it is unethical when strangers or software decide for the dead. That answer is where the research has landed. In May 2024, Tomasz Hollanek and Katarzyna Nowaczyk-Basińska of the University of Cambridge's Leverhulme Centre for the Future of Intelligence called for consent from the person whose data builds the bot, age limits, constant transparency that users are talking to an AI, ways to opt out, dignified ways to retire a deadbot, and a ban on using deadbots to advertise (University of Cambridge, May 9, 2024). One of their warning scenarios describes a parent who prepays a 20-year subscription, after which the children receive messages they cannot stop.

German research reached the same place from the grieving side. Nora Freya Lindemann argued in Science and Engineering Ethics in 2022 that deathbots can interfere with the normal course of grief, and Jessica Heesen of the Edilife project at the University of Tübingen has described digital avatars as working like a painkiller that can delay acceptance. The 2026 interview study by Li and Lan found that users value a private space for unresolved feelings and also found that invented memories start to blur with real ones. Each of those findings points to the same design rules, which are consent, control by named people, honesty about what the bot is, and a clean way to stop.

Every one of those rules is a decision a person can make in advance with Executor Lock™. You give consent by building your own Persona while you are alive. You choose the named people who may see and talk to your Persona, and nobody else can. Your family always knows they are talking with your Persona, an AI built from your memories, and your executor can ask Idy to delete your Persona entirely, which is the dignified retirement the Cambridge researchers asked for. Your release rules decide what reaches your family, and the people you named choose whether and when to open your Persona, so nobody receives messages they never asked for.

In most countries it is not illegal to make a private AI of a dead relative, but the law in 2026 is moving quickly toward two rules, which are consent from the person while alive and a named authority after death. Commercial use is already restricted in many places, and some countries now regulate griefbot services directly. The table summarizes the rules that matter most for griefbots, and the full tracker lives at deepfake and AI likeness laws by country.

China's data law contains the most useful sentence in the whole table for anyone planning ahead. Families control a dead person's data by default, unless the person arranged otherwise before death, which means a written decision made while alive outranks everyone else's. The same logic runs through the US state laws, which ask who consented, and through Germany's century-old picture law, which hands the decision to relatives when the person left none. In the United States, the federal TAKE IT DOWN Act also forces platforms to remove AI-generated intimate images within 48 hours of a victim's request (Wikipedia). For the US state picture, read digital replica laws by US state, and for the ownership question, read who owns your AI twin after you die.

A griefbot is safe when six things are true, and every one of them has to be decided before the person dies. The six come from the Cambridge rules, the Lehigh audit and the court rulings above, and the table shows how Executor Lock™ settles each one. A family cannot add any of them after the death, because the one person who could have answered is gone.

Our Terms of Service call the lock permanent and irreversible (section 8.3), and nothing in the product can change your personality, your tone or your values after the lock. New short-term memories and big life events still reach your Persona after the lock, so your daughter can tell you she got married and your Persona remembers the news. The person, though, stays exactly as you left them. For the full role of the people you name, read what a digital executor does.

A Persona is built by the person, in their lifetime, under rules they set, while a griefbot is built about a person, usually after death, under rules someone else sets. That single difference changes who decides, what the AI knows, what the AI can say and how the record ends. The table compares the two on the points families ask about most.

The Persona also shows the shape of the person's whole life, which a griefbot is not designed to do. The Mind Map arranges every memory as a constellation of chapters in a life, so a family can see which stories are there and tap any memory to talk about it. Persona Video turns your Persona into video messages in your face and your own voice, made from one 30-second recording of you, and your first video is free. For what to record on video for your family, read how to leave videos for your children.

Idy began with the questions a griefbot leaves unanswered, and only then built the AI. Those questions are who controls a person's likeness while they live, who decides what happens to the likeness when they die, who may see what, and how consent is recorded so that software actually obeys the consent. Our answer is Executor Lock™, and Idy's 78 patents pending cover that work, including verifying a death before control of a Persona changes hands, handing authority to a named executor, sealing a memory record that cannot be altered after the handover, and recording proof that consent was given. Idy has also filed two international patent applications under the Patent Cooperation Treaty, so the protection is being pursued beyond one country.

Because that foundation came first, the same system can safely run more than a private Persona. An Idy twin for working professionals answers clients as the professional while the professional lives and passes to the executor the professional named, and the same rules for every Idy twin are set out in how an AI twin is governed after the owner dies. Video-avatar tools such as HeyGen and Synthesia were built for marketing and training video, creator clones such as Delphi were built to answer an audience, and Character.AI was built for entertainment, so none of them names an executor or plans for the owner's death. Griefbot services were built for families after a death, and Executor Lock™ was built for the person before one, while consent can still be given.

You start a Persona with a living parent by sitting down together and helping your parent build their own, in their own account, in their own words. The steps below take a first afternoon and then a few conversations a month. Your parent stays in charge of every decision, which is exactly what makes the result trustworthy later.

1. Talk about the idea together first. Explain that the Persona is theirs, built by them, and that they decide who may hear their Persona and when.

2. Open your parent's own free account. The free build holds 25 memories and needs no card, and you can sit beside your parent while they start.

3. Record the first 25 memories. The Persona asks one question at a time, and most people find the first hour goes quickly once the stories start.

4. Clone your parent's voice and record a Persona Video. The Persona then speaks in your parent's own voice, and one 30-second recording lets Persona Video make messages in your parent's own face and voice.

5. Name the executor and Trusted Contacts. Your parent chooses who takes over and who may have access, and each person accepts the Executor Lock™ Agreement.

6. Set the release rules. Your parent decides whether photos, memories and live conversation are released to the family after the lock.

7. Keep going on the Mind Map. The map shows which chapters of your parent's life are covered, so the next conversation always has a starting point.

Worked example: a daughter in Columbus, Ohio, and her father. This is an invented example with real prices. Maya, 44, starts her father's free build on a Sunday and pays for a Legacy plan for him as a gifted plan at $14.99 a month, so his Persona speaks in his own voice. After a year she has paid $179.88, and her father moves to the 20-year Legacy plan for $1,299, which gives him everything in Eternal for 20 years and works out to $64.95 a year, or about $5.41 a month. Twenty years of Eternal paid monthly would cost $7,197.60, so the 20-year plan saves the family $5,898.60, and her father has named Maya as his executor and her brother as a Trusted Contact.

All prices are in US dollars. A gifted plan lets a son or daughter pay a parent's monthly Legacy or Eternal plan, while the Persona and every decision about the Persona stay with the parent. For every plan detail, read every Afterlife.ai™ plan and price, and for the building steps in more depth, read how to make an AI version of yourself.

If the person has already died, start by saving the originals, because their real recordings, messages and photos are worth more than any imitation. Copy voicemails, voice notes, videos and message threads to two drives you own and one cloud account you control, and label each file with the date and who is speaking. Ask relatives for what they hold, since the best recordings often sit on someone else's phone. Families who used HereAfter AI are learning this the hard way, and our guide to what happened when HereAfter AI shut down explains how to request recordings from that service. WhatsApp voice notes need attention first, because WhatsApp generally deletes an inactive account after 120 days, and what happens to WhatsApp when you die shows how to save each one.

Before anyone builds an AI from those files, ask the questions the law now asks. Would the person have consented, is there a written wish, and who in the family has the authority to decide? In China the law already makes the person's own arrangement decisive and gives families authority only where the person left no arrangement, and in Germany relatives decide about pictures for ten years after a death. Our pages on talking to a dead loved one with AI and how an AI chatbot of a dead person works cover the practical side in more detail.

Then make sure your own family never faces the same question. Every adult can build their own Persona today, name the people who take over and set the rules for after, so that the people who love you hear your words and never a guess at them. For a wider view of the category, read the AI afterlife industry and how people try to live on through AI.

AI cannot bring back the dead, and the griefbots that pretend otherwise put words in the mouths of people who never chose them. A US appeals court, Cambridge researchers, China's data law and New York's legislature have all landed on the same principle, which is that the person's own consent, given while alive, decides what may be made of them. Afterlife.ai™ is built on that principle, with a Persona you build in your own words and voice, executors you name and a lock that keeps you exactly as you were. Building your Persona starts free, takes an afternoon, and settles the question for your family before anyone else can answer it for you.

1. Build your own Persona free, no card

2. Start a Persona with your parent, free

3. Name your executor

We read the court, news and research sources linked in each section, including reporting from the United States, China, South Korea, Japan, Germany and Italy, the University of Cambridge and Lehigh University studies, two arXiv papers on deadbots, and the statutes and draft rules in the law table. Every outside fact links to its source and date.

• what happened when HereAfter AI shut down

• how an AI twin is governed after the owner dies

• who owns your AI twin after you die

• deepfake and AI likeness laws by country

• what a digital executor does

• Executor Lock™

• how to preserve a voice after death