Talking to a Deceased Loved One Through AI: What It Really Is, and What It Is Not

Yes, it exists: today you can hold a conversation with an artificial intelligence persona of a person who passed away, a persona built from the photos, recordings and stories they left behind, answering in their style, sometimes even in their reconstructed voice. And no, it is not what some of the headlines are selling: it is not the person themselves, not channeling and not "bringing them back to life". In this article we go honestly through what this technology is, what it is not, and what research can say about where it meets grief.

How does it work, in simple words?

Everything starts with real materials. The family collects what remains: photos and the stories behind them, voice messages, WhatsApp chat exports, memories family members tell. An AI system reads all of it and identifies patterns, how the person phrased things, what they laughed at, what mattered to them, which expressions kept recurring. From those patterns a personality profile is built, we call it a "digital DNA" and on top of it rests a persona that answers questions in the style learned from the materials.

If enough recordings were saved, the voice can also be reconstructed so the answers are not only phrased the way they phrased things, but also sound like them. We went deeper into the technology in the article What is an AI avatar.

Some people call this a "bot of a deceased person" or a "chatbot of the departed". The name is a little misleading: an ordinary chatbot answers from general knowledge gathered off the web; here the persona answers only from what the family itself collected, so it is exactly as good as the materials, and no better. Ten photos and three voice notes give a thin persona; a hundred stories and a few hours of WhatsApp give someone you recognize.

What it is not, and this needs to be said clearly

• It is not the person. There is no consciousness, emotion or soul there. There is a reflection, as accurate as the materials allow, of how the person spoke and thought.
• It is not a seance and not mysticism. The persona "knows" nothing that is not in the materials the family provided. It will not reveal secrets and will not deliver messages from the world beyond.
• It is not a substitute for grief. No technology exempts anyone from the parting. Used properly it is a tool for memory, not a way to avoid facing the loss.
• And it is not supposed to impersonate. A responsibly built persona always presents itself as AI. A platform that encourages you to forget you are talking to a computer, crosses a line.

On the wider question, whether it is right at all to build a persona of someone who can no longer consent, we wrote separately, in Digital immortality: the ethical questions.

What does research say about grief and AI?

The field is young, and honesty requires saying: there is no definitive answer yet. Studies published in 2026 found that for some mourners, a conversation with an AI persona of their loved one provided comfort and a sense of continuity. On the other hand, ethics researchers, including a University of Cambridge team that called in 2024 for regulating the field, warn against blurring the line between memory and presence, especially for those in acute stages of grief.

The bottom line accepted by most people working in the field: an aid, yes; a substitute for grief processing or professional support, no. If the grief weighs heavily, a professional comes before any avatar. We wrote about this at length in the article AI and grief.

Why do it at all?

Whoever asks "why would anyone want such a thing" usually has not yet lost someone whose voice they miss. The need is very real: to hear again how Grandma told that story; to let a grandchild born afterwards meet the person they never got to know; to ask "what would he say about this" and get an answer that feels familiar. For many families this is not an escape from reality, it is a way to keep the memory alive within it. And if the person is still here, the best way is to start now, in their own voice: that is the subject of A Voice for Generations.

Questions worth asking before you start

• Who controls the persona? The family should decide who talks with it, and be able to switch it off or delete it at any moment.
• What is it built from? Only from materials the family itself provided, not from scraping the web.
• Does it identify itself as AI? Always, in every conversation.
• What happens to the materials? Not sold, not used to train general models, deleted on request.
• Are there content boundaries? A responsible system does not discuss the circumstances of the death and does not invent memories that never were.

These, by the way, are exactly the principles we built MemoAgent on, they are laid out on our ethics page, and every family is welcome to measure us against them.

And what exists in Hebrew today?

Most of what has been written in Hebrew about "talking to the dead with AI" deals with foreign products, American and Chinese apps that made headlines. MemoAgent is an Israeli platform built from the opposite direction: the persona is built only from what the family itself gathers together, photos, recordings, WhatsApp, stories, and a living album is kept alongside it from the same materials. In Hebrew or English, from your memories, and with the boundaries we described above. You can read how it works step by step here, see a live memorial page, or start from our broad guide to digital memorialization.

FAQ