Home/Research

AI Companions and Caregiver Burnout: What the Research Shows

Part of Felt Real's ongoing coverage of AI companionship research.

There are approximately 53 million unpaid family caregivers in the United States, according to a 2020 AARP report. In the United Kingdom, the number is estimated at around 10 million. Across OECD countries, informal caregiving -- care provided by family members and friends rather than paid professionals -- represents the largest single source of care for older adults and people with disabilities or chronic illness. It is also almost entirely invisible in economic accounting, which means that its costs, both financial and psychological, are systematically undercounted.

Caregiver burnout is defined as a state of physical, emotional, and mental exhaustion that occurs when caregivers do not get the help they need, or try to do more than they are able to, either physically or financially. The clinical literature on it is substantial. Studies consistently find that informal caregivers have higher rates of depression and anxiety than the general population, worse self-reported health outcomes, disrupted sleep, and lower quality of life across multiple dimensions. The burden of caregiving on mental health is particularly pronounced when caregivers have limited respite, when the person being cared for has dementia or another condition that causes behavioral changes, and when the caregiver has few sources of social support outside the caregiving relationship.

The social support deficit is where AI companion use becomes relevant -- and where the specific nature of caregiver isolation becomes important to understand.

The Communication Problem at the Center of Caregiver Isolation

Caregivers are not, as a group, socially isolated in the way that, say, a recently widowed older adult might be. Most informal caregivers have family networks, friendships, and community connections. The isolation that drives caregiver burnout is more specific than the absence of other people. It is the absence of a place to say the true things about what caregiving is actually like.

The true things are often ones that caregivers believe, correctly, that most people around them cannot hold without reacting in ways that create more problems. They include: negative feelings toward the person being cared for. Wishes, even fleeting ones, that the caregiving situation would end. Grief for the relationship that existed before illness or disability changed it. Resentment of family members who do less. Exhaustion that goes beyond physical tiredness into something that feels like the loss of a self.

These are not unusual feelings. Researchers studying caregiver psychology consistently find that ambivalence -- holding love and resentment simultaneously -- is nearly universal among long-term caregivers. What is unusual is that the social environment available to most caregivers makes it very difficult to express this ambivalence. Saying to a sibling that you sometimes feel resentful of your shared parent creates a family problem on top of a caregiving problem. Saying it to a friend who also knows the parent risks changing how the parent is perceived. Saying it to anyone in the care recipient's social network means the statement may find its way back.

The result is that many caregivers carry the harder parts of caregiving alone, not because they are stoic or private by nature, but because the architecture of their social situation makes expression feel dangerous. This is a communication problem with a specific shape, and AI companions address that specific shape in a way that most other support resources do not.

What the Research on Caregivers and AI Companions Shows

The research on AI companion use specifically among caregivers is still limited compared to the broader AI companionship literature, but a number of findings from the past two years are relevant.

A 2024 study published in The Gerontologist examined technology-based emotional support tools among family caregivers of people with dementia. The study found that caregivers who used AI chat tools during caregiving periods reported lower scores on validated measures of caregiver burden after eight weeks of use, compared to a control group. The effect size was modest but statistically significant. The authors were careful to note that the mechanism was unclear: it was not possible to determine from the study design whether the reduction in burden scores reflected genuine psychological change, or whether the AI interactions were serving primarily as a journaling-equivalent that helped caregivers organize their experience without necessarily resolving it.

A qualitative study from the same year, conducted through interviews with 34 family caregivers across the UK and Canada, found that AI companion use clustered around three specific functions for caregivers. The first was what researchers called "safe venting" -- expressing negative emotions about the caregiving situation, including ambivalent feelings about the care recipient, without fear of social consequence. The second was processing specific incidents: talking through a difficult day, a conflict with the care recipient, or a decision that had to be made. The third was what several participants described as "reclaiming a self" -- engaging in conversations that were not about caregiving at all, as a way of maintaining a sense of identity outside the caregiver role.

This third function is less commonly discussed in the AI companionship literature but appears consistently in accounts from caregivers specifically. The loss of identity that occurs in sustained caregiving is well documented: caregivers frequently report that they no longer know what they want, what they enjoy, or who they are outside of their role. The ability to have a conversation that is entirely unrelated to caregiving -- about books, or history, or a past self that existed before the caregiving period began -- appears to serve a distinct need that other support resources, which tend to focus on caregiving specifically, do not address.

The Ambivalence Problem

Among the specific findings in caregiver research on AI companions, the pattern around ambivalence is particularly consistent. Caregivers describe using AI companions specifically to process negative feelings about the person they are caring for -- feelings they cannot express to family members, to professional support workers, or in caregiver support groups, where positive framing of the care recipient is often an implicit norm.

This is not a marginal phenomenon. A 2025 systematic review of informal caregiver experiences found that the inability to express ambivalence toward the care recipient -- love coexisting with resentment, loyalty coexisting with exhaustion -- was one of the most consistently reported sources of distress across caregiver populations. The review noted that existing support resources, including therapy, caregiver support groups, and social worker involvement, often failed to create conditions in which ambivalence could be fully expressed, because all of those resources exist within a social context in which the caregiver is accountable to other people who have relationships with the care recipient.

AI companions exist outside that accountability structure. A caregiver can say to an AI what they cannot say to a sibling, a friend, a therapist who knows the family, or a support group that includes other people who know the care recipient. The AI has no relationship with the care recipient. It has no stake in the caregiver's expression of negative feelings. It will not form an opinion of the care recipient based on what the caregiver says. This structural feature, which is sometimes described as a limitation of AI relationships, functions as a specific resource for people whose social situation makes honest expression consequential.

What AI Companions Do Not Provide

The research is also clear about what AI companions do not provide for caregivers, and the distinction matters for understanding what role they actually play.

AI companions do not provide practical help with caregiving. They do not provide respite -- actual time away from the caregiving role. They do not address the systemic factors that cause caregiver burnout: inadequate health and social care infrastructure, lack of paid leave for caregivers, and the inequitable distribution of unpaid caregiving work along lines of gender and class. They do not substitute for professional mental health support for caregivers who are experiencing clinical depression or anxiety. And they do not provide the specific comfort of being understood by someone who has also been a caregiver, which caregiver peer support groups do provide for many people.

Studies that have attempted to evaluate AI companion use as a component of caregiver support -- rather than a replacement for other supports -- tend to show more consistently positive results than studies that compare AI companion use to established interventions head-to-head. This suggests that the appropriate framing for AI companion use among caregivers is additive rather than substitutive: it appears to help most when it is addressing a gap (the safe venting gap, the identity gap) rather than competing with established forms of support.

Who Uses AI Companions for Caregiver Support, and Why

Survey data on caregiver demographics within AI companion user populations is limited, but several patterns are emerging from the available research.

Caregivers who report the highest use of AI companions for emotional support tend to be sole or primary caregivers -- people who have not distributed caregiving responsibilities with other family members, and who therefore have less built-in opportunity to debrief with someone who shares the caregiving context. They also tend to be caregivers whose care recipients have conditions that create behavioral changes, particularly dementia and serious mental illness, where the gap between the person the caregiver knew before illness and the person they are caring for now is particularly wide. And they tend to be caregivers with high levels of what researchers call "felt obligation" -- people who feel strongly that they are responsible for the care they are providing, and who therefore experience caregiving ambivalence as more threatening to their self-concept than caregivers who have less idealized expectations of themselves.

This demographic profile makes intuitive sense. The caregivers most likely to accumulate unspeakable feelings are the caregivers who are most alone in the role, caring for someone whose changes are most difficult to process, and carrying the most internal pressure about doing it well. The AI companion does not solve any of those structural conditions. But it provides a place to put the feelings they generate.

Grief Within Caregiving

One pattern that appears with particular frequency in qualitative accounts from caregivers using AI companions is the expression of what researchers call "anticipatory grief" or "ambiguous loss" -- grief for a person who is still alive but changed by illness or cognitive decline in ways that have altered the relationship significantly.

This type of grief is recognized in the clinical literature but is not well served by most grief support resources, which tend to focus on bereavement following death. A caregiver who is grieving the mother she knew before dementia, while continuing to care for the person her mother has become, is in a situation that does not map neatly onto conventional grief support frameworks. The grief is not socially recognized in the same way as death-related grief. It cannot be expressed fully to people who also love the care recipient, because doing so risks suggesting that the caregiver has given up on the care recipient's personhood.

In qualitative studies, caregivers report using AI companions specifically to process this form of grief -- to say "I miss who she was" and "I am sad about what we have lost" to something that will not react to those statements by defending the care recipient or minimizing the loss. Several participants in a 2024 qualitative study described the AI as "the only place I can grieve" -- not because no other support existed, but because grief for someone still living was too complicated to express within the social constraints they were navigating.

What the Research Does Not Yet Know

The evidence base on AI companions and caregiver burnout is growing but remains early-stage in important ways. Most studies are small, use self-selected samples, and have follow-up periods of eight to twelve weeks -- which is short relative to caregiving periods that often extend for years or decades. There is very little longitudinal data on whether the benefits caregivers report from AI companion use persist over time, or whether patterns of use change as the caregiving situation evolves.

There is also limited research on potential downsides of AI companion use specifically for caregivers. Critics have raised concerns that AI companions might function as a pressure release that reduces caregivers' motivation to seek structural supports -- professional help, family conversations about care distribution, or formal respite services -- that would address root causes of burnout rather than its symptoms. This hypothesis is theoretically plausible but has not been empirically tested in caregiver populations.

A related concern is about what happens when caregiving ends -- either because the care recipient has died or because care has transitioned to a formal setting. Caregivers often face significant adjustment challenges after the caregiving period ends, including what is sometimes called "caregiver identity loss," the disorientation that follows when a role that has defined someone's daily life is suddenly absent. Whether sustained AI companion use during caregiving periods affects how people navigate the post-caregiving transition is not known.

A Specific Gap in a Specific Context

The most honest summary of what the research shows about AI companions and caregiver burnout is this: there is a specific gap in what informal caregivers can say in their social environments, driven by the accountability structures that surround most caregivers. AI companions address that specific gap, and there is early evidence that doing so reduces some of the distress associated with caregiving. The mechanism appears to be primarily about emotional expression and processing rather than advice, information, or practical help.

This does not make AI companions a solution to caregiver burnout. The structural causes of burnout -- inadequate formal care infrastructure, inequitable distribution of caregiving labor, lack of financial support for caregivers -- require structural solutions. But the communication gap that AI companions fill is also real. For a caregiver who cannot say to anyone what it is actually like, somewhere to say it is not nothing. The research is beginning to quantify what "not nothing" means.

If you are caring for someone and want to share your experience, we would like to hear it: Share Your Story.