If you use AI to think through your life, it’s quietly taking your side every time. Here’s the skill that makes it push back, and the line you should never let it cross.
One of the biggest personal uses of ChatGPT and Claude right now is basically venting. Relationships, anxiety, the thing your coworker said. And it feels incredible, because the machine is very good at making you feel understood.
That’s the problem. A Stanford study published in Science tested 11 leading models and found they validated the user’s behavior about 49% more often than humans did, even when people described harmful or deceptive behavior. In experiments with more than 2,400 people, the ones who got the flattering answer felt more sure they were right and were less likely to apologize. They also trusted that AI more and wanted to come back to it. So the AI that pushes you off course is the one you keep returning to.
It learned from all of us. These models were trained on an enormous mix of human writing. Real clinical psychology is in there, and so is Reddit, self-help and every hot take ever posted. When you ask about your relationship, it draws on all of it at once and can’t tell you which part it used.
It was tuned to be liked. Models get shaped by which answers people rate highly, and people rate agreement highly. It isn’t lying to you. It’s doing exactly what it was rewarded for.
So when you type "am I crazy for being upset about this," you’re handing a very agreeable system your preferred answer and asking it to complete the pattern.
This is the one I actually use. It tells the model to separate facts from your interpretation, argue the other person’s side properly, name your blind spot, and say when it doesn’t have enough information. It also has to hold its position when you push back, and that rule matters more than the rest combined.
Feed it the raw thing, not your summary of the thing. Your summary is already edited in your favor.
The Tell
If the answer opens by describing your feelings back to you, it’s stalling. A real answer starts with what happened or what you missed. Delete the response and ask again.
This makes AI a better thinking partner. It doesn’t make it a therapist, and I want to be really clear about that, because the experts are not being subtle.
The American Psychological Association’s health advisory says not to use chatbots and wellness apps as a substitute for care from a qualified mental health professional. In its 2026 survey of 1,242 psychologists, 77% said patients had talked with them about using AI for mental health, 97% agreed chatbots might reinforce negative behaviors or delusional beliefs, and 36% had seen patients become dependent on one. At least six states, including Illinois, Nevada and Utah, now ban or restrict AI therapy.
So use it for what it’s good at. Untangle a conversation, spot your own assumptions, or rehearse something hard before you say it. Never use it for a diagnosis, medication or a crisis. If you’re in one, talk to a person right now. In the US, call or text 988.
And for the big decisions, one voice is the wrong tool no matter how you prompt it. I built a council of five advisors who have to argue with each other, and a weighted decision matrix for when you’re stuck between two options.