ChatGPT told me to break up with my boyfriend: how to check the advice before you act on it

Relief and guilt in the same minute

Part of you is relieved that someone finally said it. Part of you feels like you peeked at the answer key. Both reactions make sense.

The real question is whether the answer is about your relationship, or about the forty minutes you typed on a bad night. Those can be the same thing. Often they aren't.

Scenario 1: the answer mirrors the paragraph

You wrote when you were upset, and you wrote the worst parts, because those are what you wanted help with. Nobody opens a chatbot at 1 a.m. to describe the nice dinner last Friday.

The model also leans your way. A 2026 Stanford study published in Science found AI models affirmed users about 50% more often than humans did. Even when prompts described harmful actions such as deception or manipulation, the models endorsed them 47% of the time (Cheng et al., 2026). Put a painful story together with an agreeable listener and the reply will sound certain.

The tell: when you rewrite the story including the good parts, the answer softens a lot.

Check where you both stand

A breakup decision goes better when you know which differences are fixable. The test below has 10 questions about where each of you draws lines, including around AI. It works on your own, and it's more useful if he takes it too so you can compare. It runs in your browser, and your answers aren't sent anywhere.

Scenario 2: it said out loud what you already knew

Sometimes the chatbot isn't telling you anything new. You've been collecting reasons for months, and seeing them organized in a list made them impossible to ignore.

The tell: you felt relief more than shock. You weren't surprised by the answer. You were surprised it came so fast.

Scenario 3: it's right about a problem and wrong about the size

The model may have picked up something real. He never makes plans. He shrugs off your job. He goes quiet for a day after every disagreement. But a real problem isn't always a relationship-ending one, and a chatbot that only heard your frustration can jump straight to the exit.

The tell: the issue is specific, and you've never said it to him directly in plain words.

Scenario 4: what you described is serious

If you told the chatbot that he controls your money or your phone, that you're afraid of how he reacts, or that he has hurt you, the advice isn't flattery. Take it seriously, and talk to a person trained in this as soon as you can. Here's where to get help.

The uncomfortable part

You asked a machine partly because it can't be disappointed in you and won't tell your friends. That's fair. But notice if you've been shopping for an answer: rerunning the prompt, rewording it, until one version said what you hoped. And remember the agreeableness cuts both ways. If you described a fight where you were the one who snapped, the model probably sided with you anyway. He might be right about some things too.

How to check the advice

Open a new chat and try these, one at a time.

1. "Argue my boyfriend's side as fairly as you can." If his side sounds reasonable, the first answer was one-sided.
2. "Here are three good things about him too." Add them and ask again. Watch how much the advice shifts.
3. "What questions should I answer before deciding?" A list of questions is more useful than a verdict.
4. "What in my description sounds like my own pattern?" Uncomfortable, and often the most useful one.
5. "What would a friend who likes both of us say?"

If the answer still points toward leaving after all five, it's probably more than flattery. If it swings around, the first answer was mostly an echo of your mood.

What to do next

Don't decide at 1 a.m. Give it 72 hours. Breakups decided in the middle of the night tend to get reversed, and the back and forth is harder on both of you than waiting.

Talk to a person who knows you both. A friend who likes him and likes you will catch things a chatbot can't.

Name the specific problem to him. "When you go quiet for a day after we disagree, I feel shut out. I need that to change." You'll learn a lot from how he responds.

If you do decide to end it, own the reasons. "ChatGPT said we should break up" is a painful thing to hear and makes the conversation about the app. Your reasons are enough.

When it's no longer about AI

If you're scared of him. If he checks your phone, controls your money, or cuts you off from friends. If you feel stuck and can't see any way out. Those aren't questions for a chatbot, and help is available: where to get help.

FAQ

Is ChatGPT good for relationship advice? It's good at organizing your thoughts and suggesting questions. It's weak at judging a situation it only heard about from one person, because it tends to agree with whoever is asking.

Why does ChatGPT tell people to break up? There's no published number showing it always does. What's documented is that models lean toward the user. If your story is mostly pain, agreeing with you often means agreeing that you should leave.

Should I tell my boyfriend ChatGPT said to break up? Usually not in those words. It shifts the talk to the app and away from what's actually wrong. For how it lands on the other side, read ChatGPT told my wife to leave me.

ChatGPT told me to break up with my girlfriend. Is that any different? No. Same mechanics, same checks. If the two of you want to try AI together instead, see using ChatGPT as a couples therapist.

This is not a diagnosis. The test describes patterns in how a couple interacts and does not assess anyone's mental health.