Claude can sound extremely confident even when the answer is wrong, which is why this prompt is worth trying. Rather than simply telling Claude to “be more accurate,” you can set specific rules for how it should respond before the conversation even starts. The prompt encourages Claude to question your assumptions, show its confidence level, avoid automatically agreeing with you, explain its reasoning when it disagrees, and surface uncomfortable conclusions instead of hiding them. You can add it to Claude’s preferences so these instructions stay active across future chats. Of course, it won’t eliminate hallucinations or guarantee that every answer is correct. But it can make uncertainty easier to spot and reduce the kind of “yes-man” behavior that can make incorrect answers sound convincing. The goal isn’t to make Claude more argumentative. It’s to make it more useful when you want your ideas genuinely challenged and stress-tested. Would you make this your default Claude setup? �...