The Missing Pieces

Most prompts that fail do not fail on the task statement — they fail on what is missing. The task is usually clear enough. What creates revision cycles is the output format that was never specified, the tone that was assumed, the scope boundary that was never drawn, and the negative constraint that was never written. These omissions are invisible to the writer and instantly visible in the output.

1. Missing piece 1: Output format The most common omission. When you do not specify format, you accept whatever format the model determines is appropriate. That format will be reasonable. It will rarely be exactly what you need. Define the structure before defining the task. "Return a JSON array" before "summarize these ten items."
2. Missing piece 2: Tone specification "Professional" is not a tone specification. "Formal, third-person, no contractions, 8th-grade reading level" is a tone specification. The more measurable your tone requirement, the more reliably it is produced.
3. Missing piece 3: Negative constraints Negative constraints tell the model what NOT to do. "Do not recommend specific software vendors." "Do not include personal opinions." "Do not extrapolate beyond the provided data." These exclusions prevent the most common overreach failures — the model inferring that helpfulness includes things you did not ask for.
4. Missing piece 4: Scope limits Without scope limits, the model determines scope based on what it thinks a thorough response looks like. Thorough is not always useful. "Summarize only sections 2 and 3." "Address only the financial implications, not the technical ones." "Cover this in no more than three points." Scope limits produce focused outputs.

Every output format should be specified. Surprises are not features.

— FORGE