does anyone else notice gpt mostly agrees with whatever you're already leaning toward?

genuine question for people who use these models for real decisions and not just code. ive noticed that when im torn on something and i ask gpt, it kind of senses which way im leaning and reinforces it. ask it leading and you get the exact answer you fished for.

the thing that finally helped me was asking the same question to a few different models and reading where they disagree — the disagreement is the part that actually made me think. one model on its own almost never tells me im wrong.

i liked that enough that i ended up building a little thing for myself where 5 models argue it out and a separate one settles it (war table, if youre curious), but honestly even just pasting your question into 2-3 chats and comparing does most of the work!

so my real question: do you do anything to stop a single model from being a yes-man? prompt tricks, multiple models, system prompts, something else? curious what actually works for you.