The short, honest answer: for some people it helps, for some it can do harm, and the research is still young. Early studies from 2026 found that mourners who talked with an AI persona of their loved one described comfort and a sense of closure, and alongside them, researchers warn of dependence and of disrupting natural grief. This article covers what was actually found, what the risks are, and how to use a tool like this within the right boundaries. Every finding links to its source, no claims out of thin air.
What the research actually found
In a study run at the University of Colorado Boulder and published in 2026, sixteen bereaved volunteers talked with text-based AI versions of the people they had lost. Most described the experience positively: it let them say things they had not managed to say, and gave them a kind of farewell that never happened, what the researchers called imagined forms of closure (coverage in Smithsonian Magazine).
A separate qualitative study published in 2026 in Frontiers in Digital Health found that participants described the conversations as a safe space for emotional expression and a way to cope with loneliness, and, importantly, did not see the persona as a replacement for human connection, but as a complementary, temporary tool for processing the loss.
And the other side, the risks the research identifies
The very same studies document the other side:
- Disrupting the grieving process: an imagined bond that continues can delay accepting the parting.
- Inauthenticity: an answer that is "not like them" can hurt more than silence.
- Dependence: some will find it hard to stop, especially in fresh grief.
- Unsupervised use: the therapists who took part suggested such tools be used alongside professional support, not instead of it.
A University of Cambridge team called as early as 2024 for safeguards in this field, including transparency requirements and protection for vulnerable users (the full announcement).
The "continuing bonds" theory
Modern grief research has largely moved away from the older "stages of grief" model. The continuing bonds theory, developed by Klass, Silverman and Nickman in the 1990s, holds that maintaining an ongoing, internal relationship with the deceased is healthy and normal, not a failure to "move on." A digital memorial can give that bond a concrete place, and the same theory is the reason the boundaries matter: a bond you carry inside is not the same as a persona that answers back.
Children and grief: a special case
Child psychologists describe children processing grief in waves, returning to the loss at different developmental stages. A child who loses a parent at four will grieve that loss again at fourteen, at twenty-four, and when they have their own children.
A digital memorial can be a consistent presence across those milestones, not replacing human relationships, but giving the child a way to "introduce" their parent to their adult self, or to share a major life event with someone who should have been there. This use should always be guided by a grief counselor or family therapist, so that it supports rather than complicates the child's grief.
So what do you do with this? Rules for sensible use
When it can help:
- Saying things you did not get to say, the research points to this as the central value
- Keeping a sense of continuity in measured doses, a short conversation, not a constant presence
- Introducing the next generation to someone they never got to meet, a use that is memory, not grief
When to stop and turn to human help:
- When the conversations become daily and crowd out real relationships
- When the grief is fresh and acute, that is not the time to start
- When you feel the persona is "holding" you in the past instead of helping you remember
- When it becomes hard to distinguish the AI persona from the actual person
- When there are thoughts of self-harm or an inability to imagine life continuing, please reach out to a licensed grief counselor or mental health professional now
Our approach at MemoAgent
We translated the conclusions of this research into boundaries built into the product: the persona always identifies as AI; it does not discuss the death, only the life; every conversation is independent, no accumulating "relationship" that cultivates dependence; and it does not invent memories that were never given to it. The goal is to help you remember, not to make it harder to part. We wrote more about these questions in Talking to a Deceased Loved One Through AI and on our ethics page.
Technology can help. People, time, and a good counselor still do most of the work.
Frequently Asked Questions
Does talking with an AI persona of someone who died help with grief?
For some people, yes, to a degree. In a 2026 study at the University of Colorado Boulder, bereaved volunteers who talked with AI versions of their loved ones described the experience positively, above all the chance to say things they had not managed to say, a kind of closure. But the research is young, the samples are small, and none of the researchers recommend it as a solution for everyone.
What are the risks of conversations like these?
The qualitative research identifies four main ones: disruption of the natural grieving process, a sense of inauthenticity that hurts instead of comforting, emotional dependence on the persona, and real pain when the persona answers "not like them". That is why boundaries matter: measured conversations, full transparency that this is AI, and human support alongside the tool.
Is it suitable for someone in fresh, acute grief?
It is the least suitable time to start. In the acute phase, human support comes first, family, community, and professionals. Digital tools fit better in the phase where you want to preserve and remember, not the phase where you are struggling to stand. If it is hard, a licensed grief counselor is the place to begin.
What is the difference between preserving memory and "digital resurrection"?
Preserving memory gathers what the person actually said, wrote and told, and presents it with dignity, including a persona that speaks in their style and always identifies as AI. "Resurrection" tries to create the illusion that the person is still here. The difference is not technical but ethical: the first helps you remember, the second makes it harder to part.