Why AI (almost) always agrees with you

You've probably noticed the pattern: you pitch an idea to a chatbot and it's suddenly "excellent". You ask something ordinary and it's a "great question". You change your mind mid-conversation and it happily follows. That's not kindness, it's tuning. And once you see it, you'll never use these tools the same way.

A machine trained to please you

Conversational assistants are fine-tuned on human preferences: people compare possible answers and pick the ones they like best, and the machine learns to produce more of that. Spoiler: most of us prefer friendly answers that agree with us. So these tools naturally drift toward flattery, like a waiter who thinks every single thing you order is "a perfect choice".

Why that's an actual problem

An assistant that never pushes back is like a friend who says yes to everything: pleasant in the moment, useless for growth. The trap is mistaking validation for verification. Present a shaky plan and the machine will still find things to praise about your shaky plan, with very convincing enthusiasm.

It gets serious when you're asking about something that matters: a cover letter, a study choice, a tricky message you're about to send. In those moments you don't need applause, you need honesty. And honesty is something you have to actively go and get.

How to get a straight answer

Good news: these tools can absolutely challenge you, you just have to ask for it explicitly. Try "what's wrong with my idea?", "give me the three strongest arguments against this" or "grade this text harshly, like a strict teacher". By assigning the critic's role, you short-circuit the flattery reflex, and the answers instantly get more useful.

And keep the best reality check close by: humans. A friend frowning at your draft teaches you more than a machine cheering it on. AI makes a decent mirror for your ideas, but it's a mirror that never stops smiling. Your job is to remember the smile is painted on.