Is it healthy to talk to an AI version of someone who died?

This is the question people ask us most, usually phrased a little defensively, as if they expect to be told off. So let's take it seriously, because it deserves it, and because we have an obvious interest in the answer and want to be fair about that.

The short version: for most people, used in most of the ways people actually use it, yes. With real exceptions, which we'll get to.

What the grief research says

For most of the twentieth century, the goal of grief was "letting go." Then, in the 1990s, researchers looked at what bereaved people actually did and found that nearly all of them kept a relationship with the person — talking to them, keeping their things, asking what they'd think — and that this was, on the whole, associated with doing better, not worse. The idea is called a continuing bond, and it's now central to how grief is understood.

An AI companion built from your memories is one form of continuing bond. It belongs in the same category as talking to a photo, writing letters to the dead, cooking their recipe on their birthday. The question is not whether keeping a bond is healthy — it usually is — but whether this particular form helps you live more or less.

When it helps

People tell us it helps in these situations:

• Saying the unsaid. The apology, the thank-you, the news. Many people use it once, for this, and then rarely.
• Keeping them specific. Grief flattens a person into "my loss." A companion with their sayings and opinions keeps them a person.
• The hours nobody else is awake. 2am, the car, the anniversary. The people who love you are asleep; the companion isn't.
• Children and grandchildren. Someone who never met the person can ask what she was like and hear it told in her manner.
• Not being a burden. Many grieving people stop talking about the dead because they can see others tiring of it. This doesn't tire.

When it doesn't

Honestly:

• If it replaces people. The sign is not how often you use it, but whether you're cancelling on the living to do so.
• If you need it to be them. It isn't, and the careful ones will say so. If you find yourself angry at it for not being them, that's worth noticing.
• If it keeps the death from being real. Some people in very early grief use it to not-know. That's understandable and not sustainable.
• If a professional has told you to avoid it. Listen to them, not to us.
• If it's built badly. A companion that claims to be the person, or is designed to keep you hooked, can do harm. We wrote a guide to telling careful ones from careless ones.

A simple check-in

Ask yourself, after a conversation, not during:

Do I feel steadier, or more stuck?

Steadier means it's doing what a photograph does. Stuck — the same conversation, the same hour, the same refusal — means it's time to close the tab for a while, and maybe talk to a person about it. Most people are nowhere near the line and worry about it far more than they need to.

How we built ours with this in mind

Since we're the ones answering, here's what we did so the answer could be yes:

• It says it's an AI, every time you ask, and never claims to be the person.
• It won't say anything that implies it's speaking from somewhere.
• If a conversation turns to crisis, it points to real help. That can't be switched off, by us or by you.
• It never nudges you to come back, never counts streaks, never guilt-trips. You can go a year and it'll be exactly where you left it.
• You can delete everything, and it's gone — including the voice, at our provider too.

None of that makes it right for everyone. It makes it honest, which is the most we can do from our side. The rest is the check-in above, and the people around you.

AIGriefBot is free to try — build a companion and your first 25 messages cost nothing, no card. If grief feels heavier than you can carry, call or text 988 (US), or see our support page.

honestcontinuing bonds