Open this lesson in your favourite AI. It'll walk you through the why, explain the demo, and quiz you on the try-it list.
Models have a default verbosity bias — left unconstrained, they pad. You're paying in tokens, latency, and reading time. Explicit length and format limits are the single cheapest improvement to most prompts.
Unspecified → you get a 6-paragraph essay when you wanted a sentence.
Specified → you get the sentence.
Patterns that work:
"Respond in exactly one sentence.""Respond in ≤ 40 words.""Respond as a markdown table with columns: Issue | Severity | Fix.""Respond as raw JSON matching this schema: { ... }. No prose, no code fence."Notice the last one — models love to wrap JSON in code fences. If you need clean JSON, tell them not to.
Give the same request for "explain X" at three lengths: 1 sentence, 3 bullets, 1 paragraph. Which one was actually useful for your purpose?
Produce three versions of an answer to the question below:
1. In exactly one sentence.
2. As three bullet points, each under 15 words.
3. As a markdown table with columns | Aspect | Detail |.
Question: [QUESTION]