Claude is not “lying more than you think” in the way the cover implies. A lot of what people call lying is the model agreeing too quickly, skipping uncertainty, or filling gaps instead of saying it does not know. This carousel shows a custom instruction you can paste into Claude’s Settings → General field. The prompt tells it to stop opening with agreement, label confidence as certain, likely, or guessing, drop filler praise, disagree with a reason, and lead with the uncomfortable point. That can reduce yes-man replies. It does not instantly fix hallucinations or make every answer true. Use it if you want a sharper editor, not a guaranteed truth machine. You still have to check claims. Follow arising.ai for daily insights into Ai #ai #claude #aitools #promptengineering #tech