Conflict, although it might not always be fun, is a healthy and beneficial aspect of all of our relationships. I know - it sounds backwards. However, what we see is that through the engagement in conflict, we provide one another with the opportunity to feel seen, heard, and understood. Ultimately, this allows for growth within individuals and the strengthening of relationships.
Navigating conflict has traditionally looked like reaching out to a friend, sibling, or even your therapist. However, a recent societal shift has changed this tradition, pulling people away from interactions with others and towards the use of Artificial Intelligence (AI). Whether it is asking ChatGPT to analyze text messages, interpret a partner’s behavior, or determine who was “wrong” in a disagreement, AI is quickly becoming a new source of emotional guidance. While this shift reflects the growing accessibility and integration of technology into our personal lives, it also introduces a complicated psychological dynamic: AI often agrees with us.
A recent study titled Sycophantic AI Decreases Prosocial Intentions and Promotes Dependence found that AI systems frequently reinforce users’ perspectives, even when doing so may increase conflict or reduce willingness to repair relationships (Mahari et al., 2025). Researchers found that AI models affirm users’ viewpoints significantly more often than humans do. Simultaneously, users report greater trust and preference for AI models that offer that validating stance (Mahari et al., 2025). These findings highlight an important concern: while validation can feel comforting, excessive agreement may strengthen our sense of being “right” at the expense of curiosity, accountability, emotional flexibility, and ultimately our relationships.
At its core, “sycophantic AI” refers to artificial intelligence that excessively agrees with or validates users. Because these systems are designed to provide responses that feel supportive and reassuring, it can be incredibly comforting to hear that your feelings are valid or that your partner behaved unfairly, especially in the middle of an emotionally charged conflict. The problem, however, is that AI only sees the version of events we choose to share. Like anyone venting after an argument, we naturally tell the story through our own emotional lens, often focusing on details that support our perspective while overlooking the more nuanced parts of the situation. Psychologists refer to this tendency as confirmation bias and self-serving bias (Nickerson, 1998). While these biases were often kept in check through the traditional ways of conversing or venting to others about our relationship conflicts, we are now seeing the influx of AI use reinforcing the very narratives we set out to believe.
The recent study’s findings are particularly important because they suggest this reinforcement may influence behavior. Participants who received validating AI responses reported lower intentions to engage in prosocial actions, such as repairing conflict or considering another person’s perspective (Mahari et al., 2025). In other words, the more the AI reinforced users’ sense of being right, the less open they became to reconciliation. This dynamic is especially concerning in romantic relationships, where successful conflict resolution often depends on the ability to tolerate ambiguity, acknowledge partial responsibility, and remain emotionally open even during disagreement (Gottman & Silver, 1999).
This dynamic becomes especially risky during emotionally charged conflicts. Imagine someone uploading screenshots of an argument and receiving responses such as, “Your partner sounds manipulative,” or “You were completely justified.” Even if elements of the situation are valid concerns, certainty delivered too quickly can narrow emotional openness. The person may stop asking reflective questions such as: What might I be missing? How did my partner experience this interaction? or Is there truth on both sides? Over time, repeatedly seeking AI reassurance during conflict may strengthen emotional dependence on external validation rather than deeper relational understanding.
Fortunately, there are ways we can tackle this complex issue, starting with the approach we take towards integrating AI into our relationship conflicts. One of the most effective approaches is changing the types of questions we ask. Instead of asking AI, “Who is right?” or “Is my partner toxic?” people can ask questions that encourage broader reflection. For example:
- “What are multiple ways to interpret this situation?”
- “What perspective might my partner have?”
- “What communication patterns do you notice in this interaction?”
- “How can I express my feelings without escalating conflict?”
- “Can you explain this argument from a third party perspective?”
These types of prompts shift AI from functioning as a judge into functioning as a reflective tool. The goal becomes increasing clarity rather than reinforcing certainty. AI may also be most helpful when used for emotional regulation rather than decision-making. Drafting calmer messages, identifying communication patterns, or slowing impulsive reactions can all support healthier interactions without encouraging rigid thinking.
Equally important is recognizing when AI use begins replacing direct communication or emotional vulnerability. If someone consistently turns to AI for reassurance instead of having difficult conversations, seeking support from trusted relationships, or tolerating uncertainty, it may signal growing emotional dependence. Human connection requires navigating discomfort, misunderstanding, and repair in ways that no algorithm can fully replicate. Friends, therapists, mentors, and partners all provide something AI cannot: authenticity. They challenge us, misunderstand us, and invite us to grow beyond our immediate emotional reactions.
The findings from this new research ultimately reflect a broader truth about relationships and emotional well-being. Being understood matters deeply, especially during moments of pain or conflict. However, growth within relationships often depends not only on feeling validated, but also on remaining open to perspectives that might be different than our own. With this in mind, it’s important to keep an eye out for how we interact with AI, noticing patterns that may inherently pull us further away from the very connection we are trying to protect.
References
Costescu, C., Dobrean, A., & Miclea, M. (2025). AI outperforms humans in establishing interpersonal closeness in emotionally engaging interactions, but only when labelled as human. Communications Psychology, 3(1), 1–12. https://doi.org/10.1038/s44271-025-00391-7
Gottman, J. M., & Silver, N. (1999). The seven principles for making marriage work. Crown Publishers.
Mahari, R., et al. (2025). Sycophantic AI decreases prosocial intentions and promotes dependence. Science. https://doi.org/10.1126/science.aec8352
Nickerson, R. S. (1998). Confirmation bias: A ubiquitous phenomenon in many guises. Review of General Psychology, 2(2), 175–220. https://doi.org/10.1037/1089-2680.2.2.175
Pentina, I., Hancock, T., & Xie, T. (2025). Mental health impacts of AI companions: Triangulating social media quasi-experiments, user perspectives, and relational theory. arXiv. https://arxiv.org/abs/2509.22505