Can AI Help Bridge the Psychological Gap Between Patients and Their Loved Ones?

Yes, AI can help bridge the psychological gap between patients and their loved ones, but only as a bridge back to real people, never as a replacement for them. Personalized AI companions and consented “AI versions” of family members could keep familiar voices, memories and routines present during long hospital stays, serious illness or separation. The evidence is still early, so the safest uses are narrow, time-limited and supervised by the care team.

That qualification matters because the people most likely to use these tools, patients who are frightened, isolated or cognitively vulnerable, are also the people most exposed to their risks.

What are “AI versions of people” in healthcare?

Digital human healthcare tools simulate aspects of a real person: their voice, phrasing, memories or face. The best-known examples are griefbots, which build a likeness of someone who has died from their texts, emails, photos and recordings.

The hospital use case is different. Here the person being simulated is alive and reachable, and has agreed to it. A parent working abroad, a spouse who can only visit on weekends or a grandparent too frail to travel could approve an AI companion that speaks in a familiar way when they cannot be there.

That difference sets the design goal. A personalized AI companion for a patient should keep pointing back to the real relationship, not stand in for it.

Potential uses of AI for patient emotional support

The most promising uses are specific and bounded:

• Off-hours reassurance. At 3 a.m., when no one can visit, a familiar voice can offer comfort, read a bedtime story to a child or remind an anxious patient that family will call in the morning.
• Bridging distance and time zones. An AI companion can keep a gentle log of the patient’s day and help relatives abroad catch up, so the next real call starts from shared context.
• Reducing loneliness in older patients. In one experiment, AI companions with stronger relational and emotional skills reduced older adults’ loneliness, though the effect depended on how users felt about AI.
• Legacy and continuity during serious illness. Patients can record stories and messages that help families later, an extension of legacy projects already used in palliative care.
• Therapist-guided grief work. Psychologist Robert Neimeyer’s study of 10 mourners found most experienced chatbots as helpful in re-engaging with life, and Dutch researchers are testing therapist-led sessions with avatars of the deceased.

Limitations and risks

The risks are real and, in places, well documented:

• Dependence and displacement. In a four-week randomized trial of 981 people by MIT Media Lab and OpenAI, heavier daily chatbot use went with more loneliness and emotional dependence and less socializing with real people. The link is correlational, but it is the opposite of what a family-connection tool should do.
• Consent and distress. Cambridge researchers warn that AI recreations of loved ones can cause serious psychological harm when people encounter them without wanting to, and that both the person simulated and the people interacting need protection.
• Blurred reality. Patients with delirium, dementia or young age may not grasp that the voice is not really their relative, which turns comfort into deception.
• Low trust among professionals. In a survey of 53 therapists and grievers, participants largely rejected griefbots on ethical and psychological grounds, though therapists saw a possible complementary role.
• Privacy and manipulation. These tools run on intimate data, and a commercial product could use a loved one’s voice to sell or retain users.
• Errors in a loved one’s voice. An AI may say things the real person never would, including medical claims.

Safeguards and the role of clinical oversight

Clinical oversight is what separates a therapeutic tool from a risky gadget. Bioethicist Craig Klugman of DePaul University argues that griefbots should be treated like medical devices, prescribed and used only with a mental health professional’s guidance. Psychologist Robert Neimeyer, director of the Portland Institute for Loss and Transition, puts the principle simply: “give the AI a clear job to do.”

In practice, responsible use of AI for patient emotional support should include:

1. Explicit, revocable consent from the person being simulated and from the patient or their legal representative.
2. Constant disclosure. The companion always identifies itself as AI.
3. Clinical screening for delirium, dementia, severe depression or psychosis before use, with the care team able to pause it.
4. A defined purpose and time limit, plus an off switch and a plan to retire the companion respectfully.
5. Routes back to humans. The AI prompts calls with real relatives and escalates distress to staff, and gives no medical advice.
6. Health-grade data protection under GDPR or HIPAA, with no advertising or engagement targets.
7. Outcome monitoring of mood, sleep and real-world family contact