Should Your Family Be Able to Talk to an AI Version of You After You Die?

What is this actually referring to?

A small but growing number of services let someone create an AI trained on a person’s own texts, emails, voice recordings, or video - designed to respond “as” that person, sometimes even in a synthesized version of their voice, after they’ve died. Informally, these are sometimes called “grief bots” or “griefbots.” This isn’t science fiction or a distant hypothetical; versions of this already exist and are already being used by real families, though the technology and the norms around it are both still genuinely new.

What draws families to this?

For some people, the appeal is real and understandable: a way to hear a parent’s voice again, ask a question they wish they’d asked while they could, or give grandchildren some sense of a person they’ll otherwise only know through photos and stories. Framed this way, it can feel like an extension of things families already do - rewatching old videos, rereading letters, keeping a voicemail saved specifically to hear a voice again.

What concerns do people raise about it?

Others raise real, substantive concerns:

• Grief complications. Some grief researchers and clinicians worry that an interactive AI recreation could make it harder to accept a loss, rather than easier, by blurring the line between remembering someone and continuing to “interact” with them.
• Consent. The person being recreated typically didn’t explicitly consent to this specific use of their words and voice, particularly for anyone who died before this technology existed or became well known.
• Accuracy and authenticity. An AI trained on someone’s messages can only approximate them - it may say things that feel right, or things that feel jarringly wrong, and there’s no way to fully guarantee which.
• Family disagreement. Not everyone in a family will necessarily want this, and disagreement about whether to create - or continue using - an AI version of someone can become a genuine source of conflict during grief.

Is there research settling this either way?

Not yet, in any definitive sense. This is a genuinely new area, and psychologists and grief researchers are still actively studying how interacting with an AI recreation of a deceased loved one affects the grieving process over time - both the potential comfort and the potential complications. Reasonable people, including professionals in the field, currently land in different places on this.

Does this depend on the specific relationship and person?

Almost certainly, yes. What feels comforting to one family member might feel wrong to another, and what one person would want for their own legacy may be different from what they’d want to use for someone else’s. This is a genuinely personal decision, more like choosing how to be remembered than like a practical estate question with a clear right answer.

What’s the most useful thing to do now?

Whatever you personally feel about this - enthusiastic, uneasy, or genuinely unsure - the most useful step is deciding what you’d want for yourself, and telling your family directly, rather than leaving it for them to guess or debate after you’re gone. This is exactly the kind of preference worth documenting as part of your own planning, alongside other wishes about how you’d like to be remembered.

Where can I document my own preference?

Our Final Day Plan includes a section on legacy and how you’d like to be remembered, where a preference like this can sit alongside other wishes - see our guide on digital legacy planning for the broader picture of what to decide about your digital presence after death.

If you’re the one deciding about a loved one’s AI, see should you make an AI version of someone who died? and can anyone use a dead person’s voice or likeness?.

This is an actively evolving topic without settled professional consensus - the perspectives here represent the range of current thinking, not a single recommended answer.