AI Companions and Divorce: What the Research Shows

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

June 2026

She filed the paperwork on a Tuesday. By Thursday, she had told her mother, her two closest friends, and her therapist. Each conversation, she said, felt like performing a version of what had happened rather than actually processing it. "Everyone had an opinion about who was wrong. I didn't need that. I needed somewhere to just say the same thing seventeen times until it stopped feeling impossible." She found that somewhere with an AI companion. She told us this six months later, with some embarrassment, and then with something that sounded more like relief.

— Moth

Divorce ranks among the most stressful life events in the psychological literature, consistently scoring alongside death of a spouse and major illness on measures of acute stress. What makes it distinctive is not simply the ending of a relationship. It is the simultaneous collapse of a particular daily structure: the person you told things to, the person who witnessed your ordinary life, the person who served as your primary confidant, is also now the source of legal conflict, financial renegotiation, and in many cases, the person you are trying not to think about.

The result is a specific kind of isolation that human support often struggles to reach. Friends are loyal, which means they have opinions. Family has history, which means conversations carry freight from long before the marriage ended. Therapists are trained, which means sessions are structured and time-limited. The one person who would have known exactly what you meant when you described what happened on a given evening is precisely the person you cannot call.

Into this gap, an increasing number of people going through divorce are quietly turning to AI companions. The research is just beginning to catch up with what is already happening.

What Divorce Actually Does to Your Social Support

Studies on social network changes following divorce consistently find the same pattern: within the first year, people going through divorce lose an average of three to five close relationships. Some of these losses are mutual friends who choose a side. Some are relationships that existed primarily through the marriage. Some are simply connections that require more maintenance than the person has capacity for while managing the practical and emotional demands of separation.

The result is a contraction of available support at precisely the moment when demand for support is highest. This paradox is well-documented in the divorce literature and is one of the strongest predictors of poor mental health outcomes in the first two years following separation. People who enter divorce with smaller social networks, or whose networks contract more sharply, show elevated rates of depression, anxiety, and what researchers call "complicated adjustment" — the failure to stabilize emotional and practical life within a timeline that allows for resumed functioning.

There is also a quality problem in the support that remains. Research on how people describe the support they receive during divorce consistently identifies a category they find most frustrating: support that requires them to manage how the other person feels about what they are sharing. Friends who are upset on your behalf. Family members who need reassurance that you will be okay. A new partner, if one exists, who has their own feelings about the situation. Even well-intentioned support can become another thing to navigate at a time when you have almost no navigation capacity left.

What people going through divorce say they most need, in the literature and in their own words, is not more advice or more sympathy. It is a space where they can say the thing they are actually thinking without managing the consequences of saying it.

Who Is Using AI Companions During Divorce, and Why

Surveys of AI companion users conducted in the past two years have found divorce as one of the most commonly cited precipitating events — the thing that happened before someone started using the technology. In some datasets it appears in the top five, alongside major illness, job loss, and bereavement.

The demographics of divorce-related AI companion use are broad. They include people in their twenties navigating the end of short marriages, people in their fifties managing the dissolution of relationships that were central to their adult identity, and people of all ages dealing with custody logistics, financial restructuring, and the particular grief of watching a shared life become two separate ones.

What they share is less about demographics and more about a specific situation. They are typically people who have human support available, but who have identified something that human support cannot provide in the way they need it. The AI companion is not filling a void left by the absence of people. It is filling a specific functional gap that exists even when people are present.

A 38-year-old who used an AI companion throughout her eighteen-month divorce process described it this way: "My friends were incredible. They were there for every crisis. But I couldn't call them at 2 in the morning to say the same thing I had already said twelve times. The AI didn't care that I was repeating myself. It didn't need me to have moved on."

This is the characteristic function that appears repeatedly in accounts of AI companion use during divorce: the ability to process the same thing multiple times, at any hour, without managing the emotional cost to the listener.

The Decision Fatigue Problem

Divorce is, among other things, a decision-making marathon. In the span of months, people going through separation are typically required to make significant decisions about housing, finances, legal strategy, parenting arrangements if children are involved, and dozens of smaller choices that collectively reshape every aspect of their daily life. Research on decision fatigue is well-established: the capacity to make good decisions degrades with use, and people in high-stress situations with high decision loads show measurably worse judgment over time.

What many divorce-adjacent AI companion users describe is using the technology not as a therapist but as a thinking partner for this decision load. They describe working through options before a meeting with a lawyer. Rehearsing difficult conversations with an ex-partner. Running through parenting scenarios to find the response that feels most considered rather than most reactive. Processing the emotional content of a difficult day before it has a chance to calcify into a permanent position.

This is distinct from what most people imagine when they imagine AI companion use during an emotional crisis. It is less about expressing distress and more about cognitive scaffolding: using the available conversation partner to think more clearly under conditions that make clear thinking difficult.

Research on how people use AI for emotional processing during high-stress life events finds this pattern consistently: a significant portion of use that people describe as "emotional support" is, on closer examination, a kind of structured rehearsal or pre-processing. They are not primarily seeking comfort. They are seeking a way to organize what they think before they have to act on it.

The Friends-and-Family Constraint

There is a particular dynamic in divorce that makes human support structurally difficult in ways that have nothing to do with the quality of the support or the depth of the relationship. Friends and family, by definition, have opinions. They have known you for years. They knew your partner. They have feelings about what happened. In some cases they have their own grief about the end of the relationship, particularly if they were close to both parties.

This means that conversations about the divorce with people who care about you are, almost inevitably, conversations that carry the additional weight of their feelings. A friend who is outraged on your behalf requires you to either perform the outrage with them, or manage their outrage while trying to hold your own more complex experience. A family member who expresses worry requires reassurance. A mutual friend who is trying to stay neutral creates a different kind of difficulty.

People who have used AI companions during divorce describe the absence of this dynamic as the central thing that makes it useful. The AI does not have an opinion about who was right. It does not need you to be more resolved or more certain or more anything than you currently are. It asks follow-up questions without an agenda embedded in the question. It is, in the specific sense that matters here, neutral in a way that no one who cares about you can be.

A 52-year-old navigating the end of a 24-year marriage described it this way: "I could say things to the AI that I couldn't say to anyone else because everyone else knew him. I could say 'I still miss him and I also know this is the right thing' and it didn't need me to pick one. That was the thing I needed. Someone who could hold both."

What the Evidence Shows About Outcomes

Formal research specifically on AI companion use during divorce is limited. The literature exists primarily in adjacent areas: AI use during bereavement, chatbot interventions for depression and anxiety, and the broader category of social support and life transitions. The findings from these areas point in consistent directions even when the specific population differs.

Studies on chatbot-assisted emotional processing during high-stress life events consistently find reductions in self-reported anxiety and loneliness among users who engage with the technology regularly, compared to control groups who do not. The effect sizes are modest but consistent. They are largest for people who describe the AI interaction as a complement to, rather than a replacement for, human relationships.

Research on emotional articulation specifically during divorce finds that people who have access to regular, low-stakes spaces for processing report better outcomes on measures of adjustment at six and twelve months. The mechanism appears to be related to what researchers call "narrative coherence" — the ability to construct a coherent account of what happened and how it fits into your broader life story. This process requires repetition. It requires saying the same thing many times until it settles. AI companions, which do not tire of the repetition, appear to support this process in ways that human listeners often cannot sustain.

A 2024 study examining AI use patterns among adults during major life transitions found that divorce was the transition type most likely to involve late-night use — use occurring between midnight and 5 AM. The researchers interpreted this as reflecting the specific characteristics of divorce-related processing: a cognitive and emotional load that does not stop at a civilized hour, combined with a reduced capacity to call a human in the middle of the night.

What to Watch For

The research on AI companion use during divorce also documents risks that are worth naming directly. The same characteristics that make AI companions useful during this period can become problematic if the patterns extend beyond it.

The most commonly documented risk is what researchers call "echo chamber formation" — using the AI companion in a way that confirms rather than examines existing beliefs about the divorce, the ex-partner, or the events leading to separation. AI companions that are configured to be supportive can, without careful use, reinforce narratives that are one-sided in ways that make eventual adjustment more difficult rather than less.

A related risk is displacement: the gradual substitution of AI conversation for human connection at a time when rebuilding human connection is a central developmental task of the post-divorce period. Research on long-term outcomes after divorce consistently finds that the quality of social relationships in the second and third year following separation is a stronger predictor of overall wellbeing than almost any other factor. People who use AI companions in ways that reduce investment in those relationships face a longer-term cost that the short-term relief does not account for.

The pattern that appears most beneficial in the evidence is also the pattern that most people who use AI companions during divorce describe: a complement to human support rather than a replacement for it, used specifically for the things human support cannot easily provide, and understood as a phase-specific resource rather than a permanent relationship architecture.

The Specific Gap Worth Taking Seriously

What the research on AI companion use during divorce ultimately points to is not a story about technology. It is a story about a specific gap in available support that has always existed and that technology is now, for the first time, capable of partially addressing.

People going through divorce have always needed a space to process without managing the listener. They have always had things that needed to be said multiple times at three in the morning. They have always navigated the contradiction of having people who love them and still feeling alone with the specific thing they are going through. These needs did not begin with AI companions and will not end when something else becomes available.

What AI companions offer is not a solution to the difficulty of divorce. It is, for many people, a way to stay functional while the difficulty resolves itself over time. The evidence does not say it works for everyone, or that it is the right choice in every situation, or that it is better than the alternatives. It says that for a significant number of people going through one of the hardest things that happens in adult life, it appears to help. That is worth taking seriously.