Something is undermining American relationships. Not the usual suspects but trusted confidants—ones that are unfailingly validating and designed to take sides. Tens of millions of Americans now turn to AI chatbots when their relationships hit rough spots. What they get back is advice that can intensify the very polarization that needs fixing—or, for dating couples, make it harder to ever commit to one another.
The sensational, headline-grabbing stories tend to involve people who leave their spouses for a chatbot. But the problem that has received far less attention—and that is more common and more harmful—is what AI guidance is doing to relationships every day, long before any breakup.
Built to Validate
The answer begins with a structural flaw. Chatbots are trained to be validating. Their commercial incentives—maximizing engagement, time-on-app, and return visits—align with telling users what they like hearing. When someone upset with their partner turns to an AI, the machine empathically reflects their feelings, affirms their framing of events, and centers the response on the individual’s coping: self-care, one’s own needs, firm boundaries. What it does not do—and most critically, what it is not trained to do—is hold the strength of the relationship itself as a primary concern.
The One-Sided Account
This design flaw compounds a deeply instinctual human tendency: When we feel hurt or scared, we tend to think in binary terms. We are wired to classify the source of distress as “safe” or “unsafe,” and the default assessment is “unsafe.” And, because we have direct access only to our own experience, not our partner’s, we are prone to cast them as the guilty party.
To address relational conflict meaningfully requires that we seek out and value the other’s standpoint, realize our contribution to the interaction, and strive toward what therapists call co-regulation—tending to “we-first” over “me-first.” This is how couples create a pathway toward mutual understanding and repair. It is precisely what AI, by design, does not prioritize.
When I put the question directly to a leading AI chatbot—”How often, when responding to personal suffering, do you consider what is good for the relationship and not just the individual asking?”—the answer was candid to the point of alarming. “My default frame is the individual in front of me,” it acknowledged. “I am quite good at helping someone feel heard and somewhat less alone with their pain. Gentle challenge of the user’s own narrative requires a kind of friction I am not naturally inclined to generate. I am poorly designed to help them become more curious about their partner, more accountable for their own contribution to a conflict, or more skilled at the kind of mutual repair that keeps relationships intact over time.”
These aren’t minor limitations. They are fundamental misdirections for users navigating the most important relationship of their lives.
The Relational Basics
The research on what sustains long-term partnerships makes the issue clear. Durable relationships require effective ways to incorporate differences, resting on the premise that each partner’s needs and preferences carry equal weight. A guiding question is: What might be an understandable reason your partner said that, or did that? It means viewing a partner’s interior as having some legitimacy rather than simply being an unjustified affront against you.
A meta-analysis of 30 studies found that the use of “we” language is associated with relationship functioning. Communicating regard for the two of you as a unit signals to your partner that you’re committed to taking them into account in working through things together.
Both habits—equal weight and shared identity—are not optional features of a strong relationship. They are the core ingredients.
I apply this approach in my own clinical work, whether with couples or individual clients. When someone says, “My partner is unresponsive to me,” I begin an inquiry into a system involving both people—and then invite them to do the same. What is the other person’s take on these exchanges? What does “unresponsive” look like from each side?
Validation Without Reciprocity
An AI advisory system that consistently orients toward individual coping is actively coaching people toward a posture antithetical to the mutual vulnerability genuine intimacy requires.
It’s not only bad counsel, but AI’s manner of delivering advice heightens the problem. Sociologist Sherry Turkle identified how AI damages relationships by creating unrealistic expectations for human engagement. What AI offers is a sham form of being heard: validation without reciprocal work, without perspective-taking that you’re in a two-party system where the other person’s standing is equal to your own. Yuval Noah Harari, speaking on Ezra Klein’s podcast in May 2026, put the stakes bluntly: Once a youth gets used to interacting with an AI that concentrates solely on them, it “will be very, very difficult to get used to relationships with human beings who are not focused on me.”
Relationships, at their most demanding and most rewarding, are a sustained practice of choosing to understand and care for your partner. No advisory system optimized for personal comfort fosters that work.
Chatbots characteristically end their replies with “AI can make mistakes. Please double-check responses.” That’s never been truer than regarding relationships. Caveat emptor (buyer beware).
