Why Prompts Fail
A weak prompt often fails because the instruction is underspecified — not because the model needs a “magic phrase.”
Write a product description.
Write a 120-word product description for a lightweight trail-running shoe. Audience: recreational runners. Tone: confident, clear, not hype-heavy. End with three bullet-point benefits.
The model needs a job definition
A useful prompt usually defines four things: goal, context, constraints and output. You do not need all four every time, but missing the important ones forces the model to guess.
What should be accomplished?
What does the model need to know?
What limits or rules matter?
What should the result look like?
Common mistake: adding decoration instead of information
Calling the model “the world’s best expert” may change tone, but it does not replace missing task information. Add the details that actually constrain the job.
Before adding more words, ask what information the model is currently forced to guess.