Which prompting best practice did you stop following because it made results worse?
A Facebook group discussion asked which prompting best practice people dropped because it made AI results worse. One answer: stop forcing step-by-step reasoning on everything — on simple tasks it gives the model more room to talk itself into a wrong answer with great confidence. Ask for reasoning only when the task needs it, and drop the long act-as-an-expert roleplay opener: a clear task plus a concrete example of good output beats a paragraph of flattery.