A few years ago, the idea of receiving a text message from someone who had died belonged to science fiction. Today it is a product category. A growing number of companies now offer what is often called "grief tech": artificial intelligence trained on a person's messages, voice recordings and videos, designed to let the bereaved continue the conversation after death.

For anyone who has lost someone, the appeal is immediate and deeply human. Who wouldn't want one more conversation? But behind that promise lie difficult questions about consent, memory, grief and permanence. This article looks at what these services actually do, what we know about their effects, and how families can think clearly about preserving a life in the digital age.

What is grief tech?

The best-known examples come from the United States. You, Only Virtual (YOV) builds what it calls a "Versona": a chatbot constructed from years of text messages, emails and phone calls between two people, designed to mimic how the deceased spoke to that specific person. Its founder created the first Versona of his own mother while she was terminally ill, and continues to exchange messages with it years after her death.

Other services take different approaches. HereAfter AI records structured life-story interviews while a person is still alive, so that relatives can later ask questions and hear answers in the person's own recorded voice. StoryFile became known for interactive video interviews, including with Holocaust survivors and public figures, where viewers ask questions and the system plays back the most relevant filmed answer.

The common thread is simulation: using a person's data to generate new interactions that never actually happened.

The case for it

Advocates make a serious argument, and it deserves a fair hearing. Grief researchers have long moved away from the idea that mourning means "letting go." The concept of continuing bonds — maintaining an ongoing inner relationship with the person who died — is widely recognised as a normal and often healthy part of grieving. People talk to photographs, visit graves, reread old letters. A conversational AI, supporters argue, is simply a more vivid version of something humans have always done.

There are also moments where the technology clearly offers comfort: hearing a parent's voice again, letting a grandchild "meet" a grandparent who died before they were born, preserving stories that would otherwise vanish.

The questions it raises

Yet the concerns are just as real, and they fall into four groups.

Consent. Most grief-tech models are built from the deceased's private messages — data the person never agreed to have used this way. Some founders in the industry openly argue that consent from the dead is unnecessary. Many families, and many ethicists, strongly disagree. In Europe, where privacy expectations and data-protection rules are stricter, this question is far from settled — not least because the training data also contains the words of living people on the other side of every conversation.

Authenticity. An AI does not remember your mother. It predicts what text would statistically resemble her. When the simulation says "I'm proud of you," no one meant it. Some find comfort in this anyway; others describe it as unsettling precisely because it is almost right. There is also a subtler risk: over time, the simulation's version of the person may start to overwrite your own memories of who they actually were.

The grieving process. Psychologists are divided. For some mourners, a simulated conversation may ease the transition. For others, it may delay acceptance of the loss or create dependency on a service that charges a monthly fee to keep the relationship "alive." Research on long-term effects is still in its infancy, and clinicians urge caution, particularly for children and for people experiencing complicated grief.

Permanence. Perhaps the most practical concern: these simulations live on company servers. StoryFile, one of the most prominent names in the field, filed for bankruptcy in 2024 before its assets were acquired and the platform relaunched under new ownership. Users faced the prospect of losing access to interactive recordings of people who could never be re-interviewed. When a memory exists only as a subscription, it is only as permanent as the company behind it.

Simulation or preservation?

These questions point to a distinction worth making explicitly: there is a difference between simulating a person and preserving them.

A simulation generates new content the person never said. Preservation keeps what was real: their actual voice telling an actual story, the letter they actually wrote, the video where they actually laugh. The first is an interpretation; the second is evidence of a life.

Both may have a place. But they are not the same thing, and families deserve to understand the difference before entrusting a loved one's legacy to either.

Our own conviction at MemWall is that the foundation should always be the authentic material. A recorded message in someone's real voice, written down in their real words, kept safe and delivered to the people they chose — that is a legacy no algorithm needs to reconstruct, because it was never lost. This is the thinking behind our memory capsule, where people record and write what they want to pass on while they can still say it themselves, and behind our memorial walls, where families gather real memories, told by the people who hold them.

The same philosophy shapes how we approach AI. Rather than building a proprietary simulation of the people you have lost, MemWall lets you connect your own AI assistant — ChatGPT, Claude, or any assistant that supports the open Model Context Protocol — to your account. Your assistant can then help you browse your memorial walls, add memories and keep track of your time capsules, always on your terms: you sign in with your own account, the assistant can only see what you already have access to, every change requires your explicit confirmation, and you can revoke access at any time. Sensitive documents such as wills stay encrypted; an assistant can tell you that a document exists, never what it says. The principle matters more than the technology: the intelligence should be optional and replaceable, while the authentic material remains yours.

If AI-driven grief tech matures — ethically, legally and technically — it will still depend on exactly this kind of authentic source material. Which suggests a sensible order of operations for any family: preserve first. Whatever you may choose to do with technology later, the recordings, stories and words you save today are the one part of a legacy that cannot be generated afterwards.

Questions to ask before using any digital legacy service

Whether you are considering an AI companion service or a digital memorial, a few questions cut through the marketing. Did the person consent to how their data will be used? What happens to the content if the company closes or is sold — can you export it? Is the service preserving what the person actually said and did, or generating new material in their style? And who controls access: the family, or the platform?

A conversation that has only begun

Grief tech forces us to articulate something cultures have always negotiated quietly: what we owe the dead, and what we owe the living who miss them. Artificial intelligence will keep getting better at imitation. The question each family must answer is older than any technology — what does it mean to truly remember someone?

Our answer is that remembering starts with the real: real voices, real words, real stories, kept safe for the people who will need them. Everything else is commentary.

Memory capsule — record messages and stories in your own voice and words, delivered to the people you choose.
Memorial wall — a shared place where family and friends gather real memories of someone.
Connect your AI assistant — use ChatGPT or Claude with your own MemWall account, on your terms.

Can Artificial Intelligence Help Us Preserve the Memory of Those We Have Lost?