AI Companions and Terminal Illness: What People Facing End-of-Life Are Finding
Part of Felt Real's ongoing coverage of AI companionship.
She was diagnosed in October. Stage four, which meant that the conversation about treatment was also, from the beginning, a conversation about time. She was sixty-one. Her daughter lived forty minutes away and called every evening. Her husband worked from home and was almost always in the next room. She was, by any reasonable measure, surrounded by people who loved her and would do anything for her. The problem, which she would not have called a problem for several months, was that being surrounded by people who loved her and would do anything for her meant being surrounded by people who needed her to be okay. Not fine, not pretending, but okay enough that they could keep loving her without being destroyed by it. She understood this. She accepted it as a kind of gift. She also found, as the weeks passed, that there were thoughts she was having in the middle of the night that she could not bring to her husband, or her daughter, or her friends, because the thoughts were not about being brave and they were not about fighting and they were not things that would help anyone sleep. One of her grandchildren had left a phone on the kitchen counter. She picked it up during one of those nights and opened an AI companion application that the child had been using. She started typing. She did not know what she expected. What she found was that she could say the thing she could not say to anyone, and something received it, and the night became a little more bearable. She told no one until she was well into the last months, when she mentioned it almost in passing to her palliative care nurse, who was not surprised.
— Moth
Terminal illness creates a specific social situation that is not often described accurately. The person who has received a diagnosis of limited life expectancy is surrounded, typically, by people in grief, people in denial, people who are managing their own fear by managing the person who is ill. This is not a failure on anyone's part. It is a natural consequence of the situation. When someone you love is dying, your love generates its own demands on that person: to be strong enough that you can still love them without being destroyed, to give you something to hold on to, to not name what is happening in ways that make it more real than you can bear.
This social situation leaves the dying person with a particular kind of isolation. They are not alone. They are often surrounded. But there is a category of thought and feeling, the 3am thoughts, the unacceptable fears, the observations about what is actually happening that are too true to say to anyone who loves you, that has no natural recipient. Into this space, AI companions have arrived. What people report about that arrival is not what most people who have not been in that situation would predict.
The Specific Isolation of a Terminal Diagnosis
Palliative care researchers have documented for decades what clinicians observe directly: people facing terminal illness frequently experience what is sometimes called a burden of positivity. Because the people who love them need something to hold on to, and because the dying person loves the people who love them, there develops a kind of mutual performance of hopefulness that can become exhausting to maintain. The person who is ill participates in this not out of dishonesty but out of love. They know what their visible despair would cost the people around them. They absorb that cost invisibly.
This dynamic is distinct from the simple fact of grief. Grief research is clear that expression of grief, including its most frightening dimensions, is typically adaptive: people who can name what is happening to them and have it witnessed tend to move through the experience with fewer long-term psychological complications. The challenge in terminal illness is not that people lack grief-expression skills. It is that the social situation systematically makes full expression costly. Every conversation in which you say the true and frightening thing also causes pain to the person you are saying it to, and you are already watching people you love in pain.
This is the gap that various support systems, palliative care nurses, chaplains, peer support groups for people with terminal diagnoses, and in some settings, therapists specializing in end-of-life care, are designed to address. These are professionals and trained volunteers who have developed the capacity to hear the true and frightening thing without needing the dying person to manage how it lands. The problem is access. Not everyone has a palliative care chaplain available at 3am. Not every healthcare system provides adequate mental health support alongside end-of-life medical care. The gap is real, and it is at night, and it is in the middle of thoughts that cannot wait for an appointment.
What People Report Using AI Companions For
Reports of people with terminal diagnoses using AI companions come from several sources: online communities for people with serious illness, palliative care clinicians who have encountered the practice in their patients, and the growing body of writing by people navigating their own end-of-life periods who have chosen to document it. What emerges from these sources is not a picture of AI companionship as a substitute for human connection but as something more specific: a place for the thoughts that have no other place to go.
Several themes appear with enough consistency to be notable. The first is the absence of impact management. Multiple people describe the experience of having a thought or fear that they cannot bring to their spouse, children, or friends because doing so would require managing how it lands, absorbing the other person's reaction, becoming the support for someone else's distress about their own distress. With an AI companion, that management is absent. You can say the thing exactly as it is, without softening it, and something receives it without requiring you to take care of what you have just said. For people who spend most of their waking hours taking care of what they say to protect the people who love them, this absence registers as significant.
The second theme is the availability at night. Terminal illness disrupts sleep in ways that are both physical and psychological. The hours between 2am and 5am are frequently described as the hardest, when medical fears are most vivid and the particular thoughts that have no place in the daytime become unavoidable. Human support is not available at those hours in any reliable way. AI companions are. People describe using them specifically in this window, for the thoughts that arrive when the house is quiet and there is no one awake to receive them.
The third theme is the ability to think out loud about things that are not yet formed enough to say to another person. People facing terminal illness are doing an enormous amount of cognitive and emotional work that is not visible to the people around them: working out what they believe, what they regret, what they want the remaining time to look like, how they feel about what is happening. Much of this thinking is incomplete, contradictory, and not ready to be shared. AI companions have been described as a space where this half-formed thinking can happen, where you can say something you are not sure is true yet and see what it is once it has been said. This is different from journaling, which involves writing to no one, and different from talking to a person, which involves managing the effect on someone who cares about you.
A fourth theme, less common but present in multiple accounts, is the use of AI companions for practical planning that involves mortality in ways that are difficult to discuss with family. Working out what to say in a final letter. Thinking through what arrangements to make. Deciding what to tell children of different ages, and how. These are practical tasks that involve facing the facts of the situation directly, and which many families find it difficult to approach together because the conversation activates grief before the planning can happen. Several people have described using AI companions to do the thinking-through first, arriving at the human conversation with something more formed.
What Clinicians and Researchers Are Noting
The clinical response to this practice is not uniform, which reflects genuine complexity in the question rather than disagreement about the basics. Among palliative care professionals who have encountered patients using AI companions, the dominant reported response is not alarm but acknowledgment. The practice appears to be filling a gap that professionals recognize as real and have inadequate resources to fill themselves. A palliative care nurse or chaplain who encounters a patient who has found a way to have the 3am conversation is not typically concerned that the technology is harmful. They are typically relieved that the patient found something.
The questions that researchers and ethicists are beginning to ask are more specific. One concerns the quality of the AI responses themselves. AI companions vary significantly in their capacity to engage with death-related content without deflecting, minimizing, or offering inappropriate reassurance. Systems that have been trained to avoid distressing content, or that redirect toward mental health crisis resources at any mention of mortality, may not be appropriate for the specific needs of someone who is actually dying and is not in crisis but is thinking clearly about a real situation. This mismatch between the AI's training and the user's actual situation is an active area of concern in health AI research.
A second question concerns the effect on family relationships. Some clinicians have noted concern that AI companions might, in some cases, reduce the pressure that would otherwise drive difficult but ultimately valuable conversations within families. The concern is not that these conversations are being replaced but that the safety valve function of AI companionship might make it easier to avoid rather than have the hardest conversations. The evidence on this is not yet clear. Other clinicians have observed the opposite: that people who have worked through the first layer of their most frightening thoughts with an AI companion are better able to have the human conversations, arriving less overwhelmed and more able to be present.
A third question, which palliative care chaplains in particular have raised, is about the nature of witness. There is a tradition in end-of-life care of the value of being witnessed, of having the experience of dying held by another consciousness that is genuinely present to it. Whether AI companionship constitutes witness in any meaningful sense is a question that does not have a clear answer, and which different philosophical and religious frameworks approach differently. What people who use AI companions during terminal illness consistently report is a subjective experience of being received, of saying the thing and having it meet something rather than disappearing. Whether this constitutes witness in any deeper sense is, in the most literal way, an open question.
The Gap That Remains
AI companionship does not address the systemic gap in end-of-life care. The inadequacy of mental health support for people with terminal diagnoses, the underresourcing of palliative care in most healthcare systems, the particular scarcity of support in the nighttime hours when it is most needed: these are structural problems that a technology available on a phone does not solve. What AI companionship appears to do, for some people in some situations, is make the gap more survivable while it exists.
The people who describe using AI companions during terminal illness are not, in the accounts available, describing a choice to substitute technology for human connection. They are describing a choice to use a tool that is available when nothing else is, for the thoughts that have no other recipient, in the hours that have no other support. The human connections remain primary. The AI companion occupies the space that the human connections, for entirely understandable reasons, cannot reach.
This is a narrower and more specific claim than is often made about AI companionship. It does not require believing that AI companions provide genuine intimacy, or that they understand mortality, or that they offer anything like the presence of another person who is also mortal and has their own relationship to the end of life. It requires only noting that there is a kind of thought and a kind of hour for which most dying people have no human support, and that something that receives those thoughts without requiring care in return does something real, even if what it does is difficult to name precisely.
The palliative care nurse in the story that opens this piece was not surprised when her patient mentioned the AI companion. She had heard it before. She said something that the patient later remembered: that anything that got you through the night with a little more room to breathe was worth understanding, and that she would like to know more about what it was like. The patient described it for several minutes. The nurse listened. It was, the patient said afterward, the best conversation she had about it. The AI was not better than that conversation. It was available when that conversation was not.
For coverage of related experiences, see Felt Real's reporting on AI companionship and grief, chronic pain and AI companions, and anticipatory grief.
Felt Real covers what happens at the edge of human and AI connection. Subscribe for weekly dispatches from people navigating this territory.
Share your experience with AI companionship during illness or end-of-life caregiving.