PM-101 · Module 2

Specificity Is Not Verbosity

3 min read

A common overcorrection after learning that prompts need more detail is writing longer prompts. Length is not the goal. Precision is. A fifty-word prompt with every word doing specific work outperforms a two-hundred-word prompt filled with redundant instruction, restated context, and apologetic hedging. The model reads everything you write, and noise dilutes signal. Every word you add that does not change behavior is a word that competes with the words that do.

Do This

  • Use precise verbs: "Extract," "Classify," "Draft," "Compare" — not "look at" or "think about"
  • State format requirements once, specifically: "Return a JSON array of objects with keys: name, role, priority"
  • Include context only when it changes output behavior — audience, constraints, existing decisions
  • Use numbered lists for multi-part requirements so nothing is missed or ambiguous

Avoid This

  • Do not restate the task in multiple ways hoping the model picks the right interpretation
  • Do not include context "just in case" — every added sentence costs attention
  • Do not hedge: "please try to" or "if possible" weaken constraint force
  • Do not write a paragraph when a sentence will do, or a sentence when a phrase will do

The discipline here is editing. Write the full prompt, then cut every sentence that does not change behavior. What remains is the specification. If cutting a sentence would not change the output, the sentence was not earning its place. This is not minimalism for its own sake — it is signal hygiene. The model gives weight to everything in the prompt. Noise is not neutral. It dilutes the instructions that actually matter.