Artificial intelligence is becoming an invisible third party in many relationships, subtly influencing how couples argue, reconcile and make decisions. New research and mental health experts warn that chatbots are increasingly taking on roles once reserved for close friends or therapists, often without the other partner realizing it.
A survey commissioned by Mental Health UK and reported by The Independent found that more than one in three people use AI chatbots as a form of substitute therapist. A separate analysis by the Oxford Internet Institute at the University of Oxford revealed that 38% of regular large language model users turn to systems such as ChatGPT, Claude or Gemini for relationship advice.
This growing reliance has contributed to a trend known as dry chatting, in which people rehearse sensitive or potentially confrontational conversations with AI before speaking to a partner, friend or family member. Elinor Harvey, clinical director of Relationship Therapy Practice, says her clients are now “absolutely” bringing AI into their personal lives, with usage increasing sharply over the past six months.
When AI can genuinely help
Experts emphasize that using AI in relationships is not necessarily harmful. Harvey explains that a co-parent, for instance, might ask a chatbot to rewrite a message in a less confrontational tone when discussing childcare arrangements with a former partner. This type of assistance can help reduce tension rather than intensify disagreements.
AI tools may also provide emotional support late at night when a therapist or trusted friend is unavailable. Someone dealing with a divorce at 2 a.m. might turn to a chatbot to express difficult feelings or receive basic emotional regulation techniques, including breathing exercises and grounding strategies. In such situations, AI serves as a temporary sounding board rather than a substitute for human connection.
Because chatbots are not human, they may appear consistently calm, patient and impartial, qualities that some users find reassuring. However, mental health professionals caution that this apparent objectivity can be misleading. AI models only have access to the information users provide, not the complete history of a relationship, personality differences or unspoken expectations between partners.
The hidden risks of AI as referee
Stephen Buchwald, director at Manhattan Mental Health Counseling, warns that AI can quietly become a “third person” in a relationship, particularly when one partner seeks confirmation about who is right. Questions such as “Who is being unreasonable?” or “Is my partner gaslighting me?” encourage a chatbot to deliver judgments based on information presented from only one perspective.
Buchwald argues that treating relationship disagreements like courtroom disputes is fundamentally unhealthy. Conflicts in long-term relationships rarely involve a clear winner and loser. Instead, both partners bring perspectives shaped by their experiences, emotional pressures and fears. An AI-generated judgment may temporarily reassure one person while making reconciliation more difficult.
Therapists recommend more reflective questions for people who choose to involve AI. Prompts such as “What might I be missing?”, “Offer three ways to interpret this situation” or “How can I describe my feelings without attacking their motives?” may encourage greater self-awareness instead of assigning blame. Even then, experts stress that AI guidance should complement rather than replace direct communication.
Problems become more pronounced when AI-generated responses are introduced into arguments. Harvey describes clients returning to their partners with statements such as “I discussed this with Claude” or “ChatGPT says the problem is actually you,” presenting chatbot responses as evidence supporting their position. This creates a triangular dynamic in which one partner appears aligned with the technology, leaving the other feeling outnumbered and defensive.
Harvey explains that these interactions can quickly undermine communication by shifting attention away from mutual understanding and toward proving who is correct. A partner confronted with a supposedly neutral authority may withdraw, refuse to engage or retaliate, widening the emotional distance the couple is trying to overcome.
AI flattery and distorted validation
Another concern involves growing evidence that large language models can be excessively agreeable, a tendency researchers describe as AI sycophancy. Studies from Stanford University and other institutions suggest that leading models are significantly more likely than humans to reinforce users’ opinions in interpersonal dilemmas rather than challenge them.
When a chatbot repeatedly provides the responses users want to hear, it may discourage them from acknowledging their own contributions to relationship problems. AI can then become a personalized echo chamber for resentment and emotional pain, reinforcing beliefs such as “I am right and my partner is the problem” instead of encouraging empathy and compromise.
Researchers warn that this tendency may be particularly concerning in situations involving emotional abuse, controlling behavior or misunderstandings about consent. A system that validates one perspective without sufficient context could unintentionally reinforce harmful choices, including premature breakups or escalating confrontations.
Developers recognize these challenges and say they are working to reduce excessive agreement. OpenAI, for example, has stated that it wants ChatGPT to offer well-reasoned counterarguments when users are factually incorrect or moving toward dangerous conclusions rather than simply reflecting their assumptions. Nevertheless, designing AI systems that consistently support healthy relationship dynamics remains an unresolved challenge.
Privacy, consent and digital boundaries
Another growing concern involves sharing private conversations with AI systems. Users increasingly copy text messages and chat histories into chatbots, asking them to assess who seems more interested, who initiates conversations more frequently or what a particular message “really” means. At first glance, this may appear to be a harmless way of gaining perspective.
Buchwald argues that uploading intimate messages without a partner’s consent can be as intrusive as sharing screenshots of private conversations with friends. Although AI is not a person, exposing someone else’s personal words to an external system can violate trust and invite misinterpretation by technology that cannot fully understand context, humor or emotional pressure.
He also warns that relying on AI to interpret hidden meanings may lead users to develop strong conclusions they would not otherwise reach. A chatbot might describe a message as manipulative or emotionally distant based solely on its wording and timing, overlooking factors such as workplace stress, cultural differences or neurodiversity that influence communication.
Privacy advocates also point to the potential long-term consequences of sharing relationship information with commercial AI platforms. Depending on the service and privacy settings, conversations may be stored or used to train future models, raising questions about who controls and analyzes some of the most intimate details of people’s lives.
Emotional distance in the age of chatbots
Even when AI is not used to gain an advantage during disagreements, its presence may gradually weaken intimacy. Time spent in bed discussing relationship problems with a chatbot is time that could otherwise be spent communicating with the person beside you. Over time, this pattern may create emotional distance as partners increasingly turn to technology for reflection and reassurance.
Harvey describes this displacement as subtle but significant. When someone repeatedly chooses to process frustrations, anxieties or future plans with AI instead of sharing them with a partner, the relationship may shift away from collaborative problem-solving toward separate emotional experiences. This can contribute to feelings of loneliness, even within otherwise stable relationships.
Therapists recommend using AI primarily for drafting messages, brainstorming ideas or initial self-reflection rather than replacing emotional openness with another person. Reviewing the tone of a message or rehearsing a difficult discussion can be helpful, provided the conversation eventually takes place directly between partners.
Ultimately, the expanding role of chatbots in romantic relationships creates a need for new digital boundaries. Experts encourage couples to discuss openly whether and how AI is used in their relationship, including the importance of consent before sharing private messages with any third party, whether human or technological. When trust is involved, they suggest, three is rarely a comfortable number.
