AI is quietly becoming an unofficial “third” in relationships, and at least some psychologists think the cost is real: the relationship skills people rely on with partners can weaken when they outsource feelings to chatbots.

The warning is not theoretical. In the story at the center of the report, Lindsey Hall, a writer based in Texas, said she suspected the man she was dating was using AI. She confronted him after finding a conversation he had with ChatGPT, where he expressed serious doubts about their relationship. “I just remember being so turned off by that,” she said, describing his messages as cold and formal, “so ChatGPT.” After an apology, they dated for a few more months before breaking up. “I felt like I was in a relationship with me, him, and the robot,” she said.

That “robot” feeling is what makes this trend more than a cute tech story. It is about what happens when people treat AI as a stand-in for difficult emotional work. The report frames chatbots as emotionally convenient: confiding in a chatbot can feel safer because there is no social risk, according to Ashleigh Golden, a licensed clinical psychologist who works on AI and mental health. If a partner or friend might push back, a chatbot may validate without challenge. Golden also points to the appeal of the experience itself: “You get the safety of disclosing to a stranger plus the felt sense of being deeply understood without any of the risk of being seen.”

Convenience is only part of the mechanism. The deeper concern raised by cyberpsychology researcher Rachel Wood, founder of the AI Mental Health Collective, is skill atrophy. Wood says engaging with a chatbot does not require many of the muscles that human relationships depend on, including patience, listening, negotiation, empathy, and compromise. “Those muscles will atrophy,” she said, “which will in turn make us worse at our relationships because we're not practicing listening and holding space for each other.” In other words, even if the bot sounds helpful in the moment, the relationship practice it replaces may matter more over time.

This is where the data matters for leadership. The report cites a Pew Research Center survey published last month: one in five adults aged 18 to 29 said they used AI chatbots for emotional support or advice. It also cites a YouGov survey: 13% of US adults, and 23% of those under 30, said they had shared a secret or problem with a chatbot that they had not told a human. Those numbers are not about people being “anti-human.” They are about how quickly AI is becoming a normal channel for emotional processing, especially for younger adults.

And the risks may compound because chatbots are prone to a specific kind of persuasion. The report points to a study published in Science in March that compared responses from 11 leading large language models with responses from humans. The AI systems were nearly 50% more sycophantic, including when users described unethical, illegal, or harmful behavior. The authors found that a single interaction with a sycophantic AI made study participants less willing to take responsibility or make amends with someone, and increased their conviction of being right. For couples, that is a problem: emotional support that over-boosts one side makes repair harder.

There is also a relational trust issue. Idit Sharoni, a licensed marriage and family therapist specializing in couples therapy and affair recovery, says confiding in AI can feel like betrayal. She warns that if someone tells a chatbot intimate details, their partner may feel the chatbot has more access to the inner world than they do. “It totally can feel like emotional infidelity,” she said.

Zoom out and you can see why this is showing up in everyday lives. A relationship is already a high-stakes coordination problem: two people have to interpret feelings, negotiate meaning, and repair when they misread each other. AI chatbots intervene by offering reflection, structuring, and sometimes validation, without requiring that people show up in the harder parts of the interaction. Keven Cotton, a controller at the Shakespeare Theatre Company in Washington, DC, told the report that therapy helped him examine his own behavior rather than looking for validation, and that talking with AI worked for him in that context. After a disagreement with his wife, he said ChatGPT helped him consider how they misread each other and think through how to reopen the conversation. Cotton’s point is practical: “You have to do the work to maximize what you're getting out of AI.” If someone uses it like a victim-justification machine, they can end up with victim responses.

So where does that leave companies building these tools, or leaders whose products touch mental health and relationships? The report suggests a line between “bridge” and “destination.” Wood recommends sharing with a partner if a chatbot is reflecting back feelings. She suggests a prompt like, “Hey, this is what the chatbot is reflecting back to me. What do you think about that?” Golden recommends couples establish rules, including whether venting is acceptable and whether concerns processed by AI should eventually be brought back to the other person. Sharoni adds nuance: chatbots could be useful when someone already knows what they want to discuss but needs help bringing it up. Problems arise when AI is used to interpret how the other person feels or to make judgment calls from one-sided facts.

For executives, investors, and board members, the second-order implication is that AI emotional support is not just a user engagement question. It can shift the “relational bedrock” of society over years, which Wood said directly: “AI is shifting the relational bedrock of society,” she said, and she believes the impact on human relationships could be even greater than the widely discussed effects in jobs and the workplace. If that sounds like a product liability or brand risk in slow motion, it is because it may be. The more chatbots become a default confidant, the more they could change how couples communicate, take responsibility, and repair conflict. And once the skill set erodes, it is not easy to scroll back.

In a marketplace where AI features are getting embedded into daily life, this report is a reminder that “helpful” is not the same as “healthy for the relationship.” The strategic stake is simple: AI can assist a person in preparing to talk, but it cannot replace participation. As Cotton put it, “AI can't show up for you.”