AI Chatbots of the Dead Are Real
Here is how they work, the real patents and projects behind them, and the one question that matters most.
AI chatbots of the dead are real. Not a film plot, not a far-off idea. Real technology, real patents, and real families already trying it. If you have searched for this, you are probably holding a mix of curiosity and ache. That is okay. Take it gently. Here is how it works, where it came from, and the one question that matters more than any feature.
What is an AI chatbot of a dead person?
An AI chatbot of a dead person is software that imitates how someone wrote, spoke, and reacted, so that the living can keep "talking" to them after they have died. People call these tools griefbots or deadbots. You type a message, and a model answers in something close to the person's tone.
It helps to be clear about what it is and is not. A griefbot does not bring anyone back. It is a pattern, a statistical echo of the words a person left behind. It can feel uncannily familiar, and for some that is a comfort. For others it can be unsettling, or it can keep a wound open. Both reactions are normal, and neither is wrong.
The key distinction across this whole category is consent. Was the chatbot built by the person themselves, while they were alive and could choose? Or was it assembled afterward, from their data, without them ever agreeing to it? That single difference changes almost everything that follows.
How they are built
Most AI chatbots of a deceased person are built from the digital trail a life leaves behind. Roughly, the ingredients are:
Text: messages, emails, social posts, comments, and chat history, used to learn vocabulary, rhythm, and turns of phrase.
Recordings: audio or video, sometimes used to shape how the chatbot "sounds" or what stories it knows.
Social and biographical data: likes, reactions, photos, and facts about the person's life, relationships, and habits.
That material is fed to a language model, which learns the statistical patterns of how the person expressed themselves and then generates new replies in that style. The more data, the more convincing the imitation can feel. But "convincing" is not the same as "accurate." A model can invent opinions the person never held and answer questions they never thought about. It is filling gaps with probability, not memory. That gap between imitation and truth sits at the centre of the ethics.
Real examples and patents
This is where the category stops being hypothetical.
The patents
Two large technology companies hold patents pointing straight at this idea.
Microsoft, US Patent 10,853,717 B2, "Creating a conversational chat bot of a specific person," granted 1 December 2020. It describes using a person's social data, images, voice data, posts, and messages, to train a chatbot that converses in that person's personality. The patent text notes the specific person could be a "past or present" entity, which is why coverage framed it as a chatbot of someone who has died. Microsoft has said it has no plans to build a product from it.
Meta, US Patent 12,513,102 B2, "Simulation of a user of a social networking system using a language model," granted 30 December 2025. It describes a model that can generate content on behalf of a user who is absent, including when the user "is deceased." Meta has stated it has no plans to develop the technology and describes the filing as defensive.
A patent is a legal claim on an idea, not a shipping product. It is important not to confuse the two. Holding a patent does not mean a company is running this service, and reporting that says otherwise is overstating the facts.
Documented projects and journalism
The real-world building has mostly happened elsewhere. In 2016, journalist James Vlahos built a "Dadbot," a chatbot trained on recorded conversations with his dying father, later reflected in his company HereAfter AI. Researchers and journalists have since documented a small but growing "digital afterlife industry" of services that offer postmortem chat experiences. Academic work, including a 2024 study from the University of Cambridge, has examined how people use these deadbots and what can go wrong.
The ethics: consent, control, and grief
This is the part to slow down for.
Consent. Who agreed to this? When a chatbot is assembled after death from someone's data, that person never said yes. Cambridge researchers have called for designers to seek consent from "data donors" before they die, rather than recreating people who had no say.
Accuracy. A model can put words in a dead person's mouth. It can be confidently wrong about what they believed or how they would respond. There is a real dignity question in letting an imitation speak for someone who can no longer correct it.
Control and what happens to it. Who can edit the chatbot, switch it off, or keep it running? Researchers have warned about being unable to "retire" a deadbot, and about simulations that can feel like an unwanted presence, sometimes described as digital "haunting." They recommend clear consent, easy opt-out, age limits, and sensitive ways to lay these tools to rest.
Grief. We make no medical or grief-outcome claims, and you should be wary of anyone who does. Experts urge caution: comfort for one person can be a weight for another. Nothing here is a substitute for human support or professional care.
The honest summary is that the technology is ahead of the norms. The thoughtful question is not "can we?" but "did they choose this, and is it being done with care?"
The consent-first way: Afterlife AI
Afterlife AI is built the right way round. The point is to remove the hardest problem in this whole category, which is consent, by putting it first.
With Afterlife AI, the person builds their own Persona while they are alive, in their own words, with their own consent. Nothing is scraped together after the fact to impersonate someone who never agreed. You decide what is captured and how you want to be remembered.
That choice is then protected by Executor Lock. At Executor Lock, your Persona is locked: it is not changed after death, and your consent explicitly covers posthumous playback. For someone who has already died, the right answer is that any digital presence should reflect what *they* chose, not what others assemble about them. Executor Lock is how that intention is held steady.
If you also choose to preserve your voice, it is consent-based voice preservation of yourself while you are alive, locked at Executor Lock and never changed afterward. Creating your voice is free for everyone; the listening experience is the paid part, and family inherits the time you have paid for. Afterlife AI is an Australian company and Australian-hosted, and your data is treated as the sensitive information it is.
You can begin with a one-time free build budget: 60 memories and 100 conversations to shape your Persona, no card and no time limit, plus one Trusted Contact and Executor Lock setup, kept for free. Paid plans, Legacy at $14.99 per month and Eternal at $29.99 per month, come later, only if you want more.
The technology is real. The one question that matters is simple: did the person choose it? Afterlife AI makes sure the answer is yes.