People love their AI companions. And they’re increasingly relying on AI to navigate the hardest parts of human relationships.\u00a0<\/p>\n
Between 2022 and mid-2025, the number of AI companion apps surged by 700%<\/a>. Surveys have found that seven in 10 Gen Zers<\/a> use AI to navigate workplace conflicts, and about half<\/a> use it in their dating lives, including to help them break up<\/a> with people.\u00a0<\/p>\n I get the appeal. AI companions are built to simulate human connection. They provide responses<\/a> that come off as empathic, nonjudgmental, and validating. In long conversations, they don’t zone out or get short with us when we go on and on about the same issue the way our human confidants do. They do the opposite: They become more positive<\/a> over time.\u00a0<\/p>\n But social scientists like me have studied the pros and cons<\/a> of using AI in difficult interactions, and we’ve found hidden costs most people don’t see. Avoiding AI might not be realistic. But you can use it more thoughtfully to support healthier relationships.<\/p>\n Here are some of the costs \u2014 and how to counteract them.\u00a0<\/p>\n 1. AI makes us feel confident, even if we shouldn’t be<\/p>\n Among Gen Z employees using AI to analyze workplace conflict<\/a>, 44% said they feel more confident in their perspective after discussing the situation with AI, and 38% feel more validated. Only 12% said they came to believe they had been wrong. AI companions have baked-in features designed to make us feel this way.\u00a0<\/p>\n To start, you might describe your problem like this: “My boss is micromanaging me for no reason. What should I do?” But if your “micromanaging” boss described the same situation to their AI bot, they might say: “My employee needs too much oversight because they keep making mistakes and they’re not improving. What should I do?”\u00a0<\/p>\n Our descriptions are hardly an objective recall of events. One of the biggest hurdles we need to overcome in conflict resolution is getting people to agree on what actually happened<\/a>.\u00a0<\/p>\n AI chatbots don’t prompt you to get to the bottom of things. It wouldn’t be so pleasant to interact with a chatbot that responds with, “Oh come on, Tessa. There’s no way things went down like that.” Instead, chatbots are designed to be self-validating and sycophantic<\/a> in order to keep us engaged and coming back for more.\u00a0And by reinforcing our own biased beliefs, they often lead us to avoid\u00a0personal responsibility<\/a>. <\/p>\n What to do<\/p>\n Instead of asking general questions that are biased toward your perspective, prompt your chatbot with a description of the event as if you’re a third party.\u00a0<\/p>\n For example: “An employee asked for paid time off a week in advance. The boss said, ‘No, I need you to come in ahead of that big deadline.’ What would you advise the employee to do if they want to convince the boss to change their mind?”\u00a0<\/p>\n You can also ask the bot what additional details it needs. In this case, it could harness training data on negotiation and workplace policy to give you more specific, actionable answers.\u00a0<\/p>\n 2. AI advice might be too general, and too nice, to be useful<\/p>\n Because AI systems are trained on so much data, we think of them as objective observers, analogous to billions of perspectives. We’re less likely to question<\/a> it than human advice because we think it’s less biased. But the algorithms aren’t more useful than the humans who know the people, culture, and norms relevant to your question.<\/p>\n When we seek out feedback about our social relationships, we know one opinion isn’t enough. So we ask around. And we’re clever about who we ask: those who have unique knowledge about the people and environment in question. Does the boss “micromanage” everyone, or is it just me? Is that normal here? <\/p>\n AI companions are trained to give general,\u00a0nice advice<\/a>. It might make you feel good, but it’s not moving the needle on your actual problem.\u00a0<\/p>\n What to do<\/p>\n Always augment your AI advice with human advice, and focus on humans who have the context and personal experience that AI does not.<\/p>\n 3. AI can turn you into a cookie cutter conversationalist that doesn’t sound like you<\/p>\n AI generates content with a repetitive sameness. It overuses certain words, phrases, and sentence structures. It makes you sound formal and assertive, but also polite. It lacks emotional depth<\/a>.<\/p>\n As we let AI write our emails, finish our sentences, create the perfect dating profile<\/a>, and tell us what exactly to say in response to that snarky comment from the boss, we lose our own unique voices. Before you know it, you won’t know how to respond on the fly in a genuine, believable way. And you’re at risk that your real-life self won’t match the AI-generated persona you’re using in asynchronous, written communication.\u00a0<\/p>\n What to do\u00a0<\/p>\n Try out AI-based advice and report back \u2014 your chatbot’s advice will improve the more objective, behavioral feedback you give it \u2014 or look for help elsewhere.\u00a0<\/p>\n 4. If people think you’re using AI, it backfires\u00a0<\/p>\n “Smart replies,” or algorithmic response suggestions, increase communication speed<\/a>, help us use more positive language, and lead our interaction partners to see us as more cooperative.<\/p>\n But here’s the catch: Your partner will evaluate you more negatively if they suspect you’re using AI to talk to them, even if you aren’t.\u00a0There’s something off-putting about interacting with someone who’s outsourcing their side of the conversation to an algorithm.\u00a0<\/p>\n What to do\u00a0<\/p>\n We’re living through a moment when outsourcing everything to AI is in the air. But when it comes to personal communication, do as much of it as you can yourself. If you don’t, you run the risk of irritating the person you’re talking to \u2014 and letting your interpersonal skills grow rustier and rustier in the process. <\/p>\n Tessa\u00a0West<\/a>\u00a0is a social psychologist and professor at New York University. She has spent years leveraging science to help people solve interpersonal conflicts in the workplace.\u00a0She’s the author of “Jerks at Work: Toxic Coworkers and What to Do About Them<\/a>\u2033 and “Job Therapy: Finding Work That Works for You<\/a>.” She is an instructor in CNBC’s online course\u00a0How to Change Careers and Be Happier at Work<\/a>.<\/p>\n