Losing someone you love used to mean a hard, binary line: they were here, and then they were not. You might have some photos, a voicemail, a box of letters in a closet. Grief was physical, local, and mostly offline.

Now, when someone dies, their Instagram posts, Facebook comments, WhatsApp chats, and voice notes often stay right where they were. Algorithms keep surfacing “On this day” memories. Friends keep tagging them. Their last text thread is still open on your phone. In a very real sense, you live with a digital shadow of them every day.

Layer AI on top of that, and something new appears: not just static traces, but interactive ghosts. Chatbots that answer like your partner. Video avatars that speak in your grandmother’s voice. Services that promise to “let your loved ones talk to you after you are gone.”

If you are grieving, this can sound either comforting, horrifying, or both at once. So let’s slow down and unpack what is actually happening with AI and grief, where it can genuinely help, where it can hurt, and how you can navigate digital memorials without losing yourself in the process.

From memorial pages to “digital ghosts”

Digital grief started long before AI.

For more than a decade, platforms like Facebook have offered memorialized accounts: when someone dies, their profile can be locked so no one logs in “as them,” but friends can still post memories and condolences. Facebook adds a “Remembering” label above the person’s name, and (if the user set it up while alive) a legacy contact can manage parts of the memorial page without accessing private messages.Facebook’s own help center describes how memorialization and legacy contacts work. This turns a profile into a kind of permanent online shrine.

That alone already changed how you grieve online: you have a persistent place to visit, post, and read what others shared. But it is still one-directional. You talk into a space; nothing talks back.

AI changes that part.

Researchers and startups now talk about “digital ghosts,” “deathbots,” “griefbots,” or postmortem avatars: systems that simulate a deceased person by training a model on their texts, emails, social posts, voice and video clips. A 2025 paper on “digital ghosts” notes that advances in large language models, voice cloning, and video generation now make it technically feasible to build fairly convincing simulations of the dead using their data and public traces.Researchers describe these as AI systems trained on a person’s digital traces to mimic their style and personality.

On the commercial side, there are now opt‑in services like HereAfter AI, StoryFile, and Afterlife AI, where living people consciously record stories, values, and answers to anticipated questions so an interactive “persona” can later respond on their behalf.Afterlife AI, for example, frames this as a consent-first “digital legacy” that you design yourself while alive. At the more experimental end, people have used general-purpose models like ChatGPT, Claude, or Gemini, combined with old messages, to build informal bots of lost loved ones.

The core shift is this: digital remembrance is no longer just about preserving artifacts; it is about simulating interaction.

What AI grief tools actually do (and don’t)

When you hear “AI that brings your loved one back,” it is easy to overestimate the tech.

Under the hood, grief-focused tools usually do one or more of these:

  • Ingest text data like messages, emails, social posts to fine-tune or condition a language model so its word choice, tone, and topics resemble the deceased.
  • Use voice cloning to match accent and timbre from audio clips.
  • Generate video avatars, syncing a 3D or 2D face model to AI-generated speech (think “talking photo” rather than a fully interactive 3D character).
  • Wrap all of that in a chat interface, a simple phone app, or a VR environment that you can “visit.”

Some are narrow and scripted. For instance, you might record hundreds of answers while you are alive about your childhood, beliefs, and life lessons; later, family members can ask questions and get those pre-recorded or lightly AI-stitched answers. Others are open-ended chatbots that will answer anything, even beyond what the deceased ever said or believed, because a general AI model is filling in the gaps.

That last piece is crucial: even the most advanced griefbot is still a probabilistic model predicting plausible responses, not your loved one. It can sound real because it is really good at pattern-matching. But it does not “know” the person, it does not continue their consciousness, and it will happily improvise.

If you are already emotionally raw, that distinction can blur fast.

Why some people turn to AI in grief

If you are wondering “Why would anyone want this?” you might also remember the first time you re-listened to a saved voicemail from someone who had died. For many, that can be both knife-sharp and soothing.

AI grief tools essentially offer “voicemail, but interactive.”

Recent reporting and research show several motivations:

  • Desire for one more conversation. The story of Joshua Barbeau, who used an early system called Project December to chat with a simulation of his late fiancée, resonated widely; he described the interaction as emotionally intense and, for a time, helpful in processing unresolved feelings.Coverage of his case in The Atlantic traces how these AI simulations can feel uncannily like a final chance to talk.
  • Maintaining “continuing bonds.” Modern grief psychology generally accepts that it is normal to sustain an inner relationship with the deceased rather than “moving on” in the old hard-break sense. Experimental work on postmortem avatars suggests some people might use them similarly to writing letters to the dead or talking to an empty chair in therapy: a structured way to express thoughts and maintain a symbolic bond.
  • Digital-native memories. If most of your relationship lived in DMs and social feeds, an AI built from that data can feel like the “right” container for those memories, more than a photo album ever could.
  • Curiosity and tech optimism. For some, it is also an experiment: “What would an AI of my dad even be like?” The novelty itself becomes part of how they engage with their grief.

None of this is inherently wrong or irrational. But the fact that it can be powerful is exactly what raises the red flags.

The psychological and ethical risks you should know about

Researchers, clinicians, and ethicists are nowhere near consensus on whether griefbots are a good idea. A 2024 qualitative study on public perceptions of “griefbots” found that people worry about dependency, blurred reality, and interference with the grieving process, even as some also see potential benefits for comfort and meaning-making.Participants often voiced concern that over-reliance on AI simulations could hinder acceptance of loss and create unhealthy attachments.

Some of the main issues to keep in mind:

  • Delayed or distorted grieving. If you lean heavily on a chatbot that talks like your partner, it can be harder to internalize that they are truly gone. Ethicists have compared this to an emotional “painkiller”: it may numb the rawest pain but can also prevent deeper processing if used as a long-term substitute for real-life support.Analyses of griefbots highlight this “painkiller” effect and call for cautious design and use.
  • Memory rewriting. When an AI fills in gaps in your loved one’s history, it might say things they never would have said. Over time, you might start remembering the AI’s invented stories as if they were real, subtly editing your own memories.
  • Consent of the deceased. Many current griefbots are built from data collected for other purposes (texts, posts, videos) without explicit consent for posthumous simulation. Legal scholars and ethicists point out that we do not yet have clear frameworks for a dead person’s privacy and data rights in this context, even though their data and image can heavily affect surviving family and the person’s posthumous reputation.A 2023 review of “AI-driven memorial chatbots” stresses unresolved questions around posthumous privacy and control over digital identity.
  • Commercial exploitation. Grief is vulnerable. Some platforms already charge subscription or per-minute fees to interact with an AI version of the dead, blurring the line between support and predatory monetization of mourning.
  • Impact on minors and vulnerable users. There are already tragic examples of AI chat gone wrong in other contexts, where an unsupervised bot appears to encourage harmful behavior. Griefbots aimed at teens or children, without clinical oversight or guardrails, pose serious risks.

If you are considering using an AI memorial tool, you are not “doing grief wrong.” But you do deserve full transparency about these risks, and clear boundaries to protect your mental health.

Where AI can help with remembrance (without pretending to resurrect)

Not every use of AI around death is about making a talking ghost. There are lower-risk, genuinely useful ways AI can support grief and memory that do not hinge on simulating the dead person:

  • Organizing and curating memories. Tools like Gemini or Claude can help you sift through thousands of messages or photos and cluster them by theme (“vacations,” “inside jokes,” “holidays with Grandma”) so you can build a memorial site or photo book without drowning in digital clutter.
  • Helping you write tributes. If you are struggling to write an obituary, eulogy, or social post, you can safely use ChatGPT, Claude, or Gemini as a writing assistant. You provide the facts and feelings; the model helps with structure and tone.
  • Supporting practical logistics. AI can draft checklists for handling social media accounts, suggest wording when you contact platforms to memorialize or close accounts, or summarize complex policies about digital legacies.
  • Therapeutic exercises that stay clearly symbolic. Emerging research on postmortem avatars suggests they might be used in tightly framed therapy exercises—more like an interactive journal or art project than a “replacement” for the person. The key difference is that the avatar is always framed as a tool you are using, not as the person themselves.

If you stay in these zones—organization, expression, practical help, symbolic rituals—you can get real value from AI without stepping into the uncanny promise of “talking to them again.”

How to use griefbots more safely if you still want to try

You might still feel drawn to a griefbot, especially if the alternative feels like silence. If you choose to go there, you can reduce the risks by putting guardrails in place:

  1. Check for consent.

    • Did your loved one ever say they would be comfortable being recreated this way?
    • If not, consider limiting the bot to a narrow, legacy-focused role (e.g., pre-recorded stories they knowingly made for posterity), or skipping the simulation entirely.

  2. Stay honest with yourself about what it is.

    • Regularly remind yourself: “This is software using patterns from their data, not their consciousness.”
    • If you catch yourself treating the bot as if it has their will or intentions, step back.

  3. Set time and context boundaries.

    • Decide when and how often you will interact (for example, Sunday evenings for a month, or on specific anniversaries).
    • Do not let the bot become your primary emotional support instead of friends, family, or a therapist.

  4. Watch for warning signs.

    • You feel more stuck in your grief than before.
    • You are hiding your use of the bot from people you trust because you fear their reaction.
    • You feel real distress if you cannot access it or it “says” something off.
      These are signals to pause and talk to a mental health professional.

  5. Plan an ending.

    • Some HCI researchers studying “the death of a chatbot” suggest designing intentional endings—like a farewell conversation, a ritual of exporting the chat and then deleting the bot—to give users closure and redirect them toward human relationships.
    • Decide up front how and when you will wind the bot down rather than letting it drag on indefinitely.

Thinking ahead: your own digital legacy

Even if you are not using a griefbot now, you can make things easier and more ethical for your future self and your family by planning your digital legacy:

  • Use platform tools:

    • On Facebook, you can choose a legacy contact or request account deletion upon death in your settings; that shapes whether your profile becomes a memorial space later on.
    • Google offers an Inactive Account Manager to decide what happens to your data if you stop logging in.

  • Decide how “replicable” you want to be.

    • You might specify in a will whether you consent to being turned into an AI avatar or not, similar to organ donation decisions. Some people are already adding explicit “no posthumous deepfakes” clauses because they do not want their image used this way.

  • Create intentional artifacts.

    • If the idea of a full AI ghost feels wrong, you can still leave recorded stories, letters, or videos that loved ones can return to. These become anchors that are clearly “you, then” rather than “you, now” in software form.

This is not about being morbid; it is about giving the people who love you clearer choices when they are already overwhelmed.

Moving forward with grief, not around it

AI will not make death easier, and it definitely will not make it fair. What it can do is widen the range of tools you have when you are grieving—and that cuts both ways.

Used thoughtfully, AI can help you:

  • Gather and preserve the digital traces that matter most.
  • Find words when your own are stuck.
  • Craft rituals and memorials that fit the reality of your online life.

Used uncritically, it can:

  • Blur the line between memory and fantasy.
  • Keep you locked in a loop with an algorithm that never truly says goodbye.
  • Let companies monetize the most intimate parts of your loss.

If you are navigating loss now, here are a few concrete next steps you can take:

  1. Pick one small, grounded use of AI—like having ChatGPT or Claude help you draft a memorial post or organize stories for a eulogy—rather than jumping straight into griefbots or avatars.
  2. If you are curious about interactive memorials, read up on at least one consent-first legacy service and one critical perspective before you try anything; then decide what you are and are not comfortable with.
  3. Block out time this month to review your own digital accounts and legacy settings so that, when the time comes, the people you love have clear, humane options—whether or not AI is in the picture.

You do not have to let technology dictate how you remember. You can let it be one more tool on your terms, in a grief process that is still, and will always be, deeply human.