Section 1 of 1
The Art of Telling AI What NOT to Do
Research shows that LLMs adhere to negative constraints (what NOT to do) with 30-40% higher compliance than positive instructions. This section is your most powerful tool for quality control.
Start with Output Inspection
Generate output WITHOUT negative constraints. Read it carefully. Circle every unwanted behavior.
Write a Constraint for Each
For each circled issue, write a specific negative constraint. Be explicit — 'don't ramble' is vague; 'maximum 3 sentences per paragraph' is precise.
Test the Constraints
Re-run the prompt with constraints. If the issue persists, make the constraint STRONGER with capitalized emphasis (NEVER, ABSOLUTELY DO NOT).
Stack at the Bottom
Place your most important constraints at the very bottom of the prompt — LLMs pay most attention to the beginning and end of context windows.