Use a self-improvement loop: generate → self-critique across named dimensions → revise, 2–3 passes

Prompt1001·medium signal

Rather than one-shot output, structure the prompt so the model produces a draft, systematically identifies its own weaknesses against specific named dimensions, then rewrites to fix them — looping two or three times. The explicit dimension list (not generic 'improve this') is what makes the critique pass actually catch errors instead of cosmetically rephrasing.