If You Are Thinking About Talking to an AI Version of Someone You Lost

It is an AI chatbot trained on materials from the deceased person. Depending on the product, the materials might include their text messages, social media posts, voice recordings, video recordings, emails, or interviews they recorded while alive. The chatbot generates responses in a style approximating the deceased person's voice and views.

What it is not: it is not the person. It is a generated approximation based on patterns in the input data. The quality of the approximation depends almost entirely on the quality and consent of the input materials. A chatbot built from scraped social media posts produces shallow, generic responses. A chatbot built from extensive interviews recorded by the person themselves produces something deeper, but still not the person.

Grief researchers have written extensively about what helps and what hurts in the weeks and months after a loss. The consistent findings: relationships with other living humans matter most. Talking to a therapist who specialises in grief, joining a peer support group, staying connected with friends and family, working with a clergy member or spiritual advisor, taking care of physical wellbeing through sleep, food, and movement. These are first-line resources.

Technology, including AI versions of the deceased, is downstream of those. For some people, in some contexts, it adds something. For others it gets in the way. The honest answer is that the technology is too new for confident claims either direction.

Dr. Jessica Heesen, lead ethicist of the Edilife project at the University of Tübingen, has described AI versions of the deceased as potentially acting like a painkiller, preventing the bereaved from accepting and processing the loss.

Nora Freya Lindemann's 2022 paper in Science and Engineering Ethics, drawing on theories of internet-scaffolded affectivity and grief, argued that deathbots (her term) may have a negative impact on the grief process and therefore limit the emotional and psychological wellbeing of their users.

The Hastings Center has covered these technologies as raising serious questions about wellbeing, while acknowledging that cultural framings vary. Some Latin American traditions, Día de los Muertos for example, integrate continued engagement with the dead in ways that may make AI interactions feel less unsettling. Some Asian traditions of ancestor veneration similarly differ from Western frameworks. The same technology may feel different in different cultural contexts.

What the literature converges on is that these technologies should not replace human grief support, should be approached with care, and should never be used by children.

Used carefully, an AI version of someone who died can do several things. It can let you revisit recordings the person made, organised around questions you might ask. It can help you access stories and context that you have forgotten but that exist in the underlying material. It can give you a kind of structured visit with what the person chose to leave behind.

What it cannot do: it cannot grow. It cannot say things the person did not record or did not want recorded. It cannot replace the relationship. It cannot tell you what they would have thought about your life today, because they did not live to see it.

There are essentially two kinds of products in this category, and they have very different ethics.

Posthumous reconstruction. The deceased did not build it. Family members or a service provider build it after death, often using scraped social media data, archived messages, or whatever else is available. This is the form most academic ethicists have raised concerns about. Consent from the person being simulated is typically not present.

Consent-first preservation. The person built it themselves while alive. They chose what to include, who could access it, what permissions would apply after death. This is the form Afterlife AI™ provides. It is structurally different from posthumous reconstruction because the consent is explicit and the creator is the data donor.

Both forms use similar underlying AI technology. The ethical difference is consent, not technical.

Afterlife AI™ is a consent-first digital legacy platform. The Persona is built by the person it represents, while they are alive. It captures who they were across eleven dimensions of identity, with the person choosing exactly what each dimension contains.

Executor Lock™ provides the governance layer that posthumous reconstruction products typically lack. The person sets, in advance, who can access the Persona after their death, under what rules, for how long. When the lock activates, the Persona transitions to read-only governance. It cannot be edited, cannot be expanded posthumously, and cannot be used in ways the creator did not approve.

What grief actually needs is a person to talk to. The technology comes later, if it comes at all.

The structure produces something honest. Not a synthetic version of the person, but a structured preservation of what they chose to share, accessible under the terms they set.

This is the painful case. The technology that exists today, with consent, is best used when the person built their Persona themselves. If they did not, there is no consent-first Persona of them to talk to.

What you can do: hold the recordings, photos, messages, and documents you do have. Revisit them at your own pace. Talk to other people who knew them. Consider working with a grief therapist who can help you integrate the loss without rushing it. The relationship continues, in the form of memory and influence, even without a chatbot.

And, if you are reading this for yourself, while alive: this is the case for building your own Persona now. Your family cannot consent on your behalf to a posthumous reconstruction. What you build while alive is what they will inherit.

This is the painful case. The technology that exists today, ethically and at quality, requires the person to have built the system themselves while alive. If they did not, there is no consent-first Persona to talk to.

Several other paths can help. Hold the materials they did leave: photographs, recordings, written correspondence, voicemails. Many people find that revisiting these materials at their own pace, without trying to construct an AI version, provides what they actually needed: a sense of continued relationship through what was real, rather than what is generated.

Talk to other people who knew them. Family members and old friends often hold stories, perspectives, and observations that the bereaved person never heard. Structured family interview projects, sometimes facilitated by life-story professionals, can produce a rich record of the deceased that no AI reconstruction could match.

Consider working with a grief therapist who can help integrate the loss without rushing it. The relationship continues, in the form of memory and influence, even without a chatbot. Most people who lose somebody discover, over months and years, that the dead person remains present in their thinking, their decisions, and their sense of who they are. This is normal, not pathological.

For families where the person did build a Persona, the experience is meaningfully different from interacting with a posthumous reconstruction.

Specificity. A Persona built across the eleven dimensions of identity captures things that a posthumous reconstruction from scraped data cannot. The person chose what to include in each dimension, often spending hours on questions they would never have answered on public social media.

Authenticity. The Persona's responses come from materials the person authored, not from patterns extracted from external observation. The voice is the person's voice, not an inferred approximation.

Governance. The rules under which the Persona operates were set by the person themselves. Their wishes about who can access what, when, and under what conditions, are enforced by Executor Lock™ rather than negotiated between family members after their death.

Honesty. A well-built consent-first Persona is honest about what it is. It does not pretend to be the person. It is a structured preservation of who they chose to share, accessible under terms they set, useful in the ways they intended.

The case for building your Persona while alive becomes most clear when you imagine the alternative for your own family. If you die without building one, the only way for your family to interact with an AI version of you is posthumous reconstruction, with all the ethical concerns that involves and without any of the consent and governance that make the form work.

Building a Persona is not for everybody. Some people prefer a different kind of legacy: written letters, recorded interviews, traditional memorial practices. These have their own value. But for people who want their family to have an interactive way to remember and learn from them, building during life is the only ethically uncomplicated path. The work cannot be done after death by anybody else.

The most rigorous public-facing examination of what it is actually like to talk to an AI version of a deceased person came from researchers at King's College London and Cardiff University, publishing in the academic journal Memory, Mind and Media in late 2025 and writing up the findings for The Conversation. The researchers, including Eva Nieto McAvoy at King's College London, became their own test subjects. They uploaded videos, voice notes and messages of themselves to multiple commercial deadbot services and then tried to interact with the resulting digital doubles.

The findings: the conversations felt flat. Stiff, scripted replies. Cheerful emojis appearing alongside questions about death. The more personalisation the researchers attempted, the more artificial the responses felt. The researchers described this as the limits of synthetic intimacy. The business model behind the experience also drew critique: subscription fees, freemium tiers, partnerships with insurers and care providers, all of which translate remembrance into a commercial product.

James Muldoon (Associate Professor in Management, University of Essex) examined a different angle in The Conversation in January 2026, drawing on his book Love Machines. Muldoon described the case of Roro, a content creator in China whose deceased mother became a public chatbot on the Xingye platform. The piece illustrated how a single posthumous AI representation, built without explicit consent during life, can become public infrastructure shaped by people other than the deceased and their immediate family.

The pattern across the academic and press coverage is consistent: talking to an AI version of a deceased person feels different depending on whether the deceased had a voice in how the representation was built. Tom's Guide writer Jason England, in his February 2026 piece, drew the same line between opt-in services such as Afterlife AI™, StoryFile and HereAfter AI and automated approaches described in patents such as Meta's US12513102B2. The opt-in services let the person being preserved set the rules. The automated approaches do not.

Patrick Stokes, Associate Professor of Philosophy at Deakin University and author of Digital Souls: A Philosophy of Online Death (Bloomsbury, 2021), drew the clearest distinction available in the academic literature when asked by the Daily Telegraph in January 2026 what people are actually talking to when they interact with an AI version of a deceased person. With a phone call, you are connecting to another consciousness. With a bot, you are not, you are connecting to a prediction machine that just works out what the next line would sound like in a real conversation. The clarity of that distinction is why his book has become a standard reference in the philosophy of online death.

The implication for anyone considering whether to talk to an AI version of someone who has died is that the experience is a kind of guided reflection on the deceased, not a conversation with them. The reflection can be valuable. It is not the same as the deceased. Stokes's other concern in the Telegraph piece, that society may stop caring about the difference between synthetic and real people, is the cultural risk if this distinction becomes blurred. The honest framing for any service in this category is that the Persona is a structured record of who someone chose to be, played back through a probabilistic prediction layer. It is not them. The decision to talk to it should be made with that understanding.

Can I build an AI version of someone after they have died?

Technically yes; ethically the answer is much more complicated. Most academic ethicists oppose posthumous reconstruction without explicit prior consent. Afterlife AI™ does not offer this. The Personas on the platform are built by the person they represent, while alive.

Will it really feel like them?

Partially, at best. The technology generates approximations, not the person. Many users report a mixed experience: moments of recognition followed by moments of clear difference. The quality of the experience depends heavily on the quality and consent of the source material.

Is it bad for grief?

It might be. The research is too early to be confident either way. Most ethicists recommend using these technologies cautiously, alongside (not instead of) human grief support, and never with children.

What does Afterlife AI™ offer that other companies don't?

Consent-first design. Personas are built by the person they represent while alive, governed by Executor Lock™ under rules the creator set. This is structurally different from posthumous reconstruction.

Should I do this if I am still actively grieving?

Talk to a therapist first. If you are in early or acute grief, technology should be downstream of human support, not a substitute for it.