1. The four-part skeleton

Most weak prompts are missing one of four things. TASK: what exactly should exist when this is done? AUDIENCE: who is it for, and what do they already know? CONSTRAINTS: length, tone, what to avoid, what must be true. OUTPUT SHAPE: email, table, bullet list, three options, one recommendation?

'Write something about our pricing change' fails all four. 'Write a 120-word email to existing customers announcing the March pricing change, plain, direct, no marketing language, end with the support link' answers all four in one sentence.

2. Specific beats long

Clarity is not word count. A 300-word ramble can be vaguer than a two-line ask. The question to ask yourself: could two different people read my prompt and produce two very different things? If yes, it is underspecified, no matter how long it is.

Precision usually costs a handful of words: a number ('under 150 words'), a name for the reader ('for the board'), a named enemy ('no jargon'), a shape ('as a two-column table').

3. Verbs do heavy lifting

'Summarize', 'critique', 'rewrite', 'compare', 'brainstorm' and 'draft' are different jobs that produce different outputs. Choosing the accurate verb is the cheapest instruction you can give. Asking for a 'review' when you want a 'rewrite' guarantees disappointment.

Stack verbs deliberately when the job has stages: 'First list the three weakest arguments, then rewrite only those paragraphs.' That is a workflow in one sentence.

4. Diagnose by the missing part

When an answer disappoints, resist the urge to re-roll. Diagnose: wrong content usually means the TASK or AUDIENCE was unclear; wrong level of detail points at AUDIENCE; wrong style or length points at CONSTRAINTS; wrong structure points at OUTPUT SHAPE.

One targeted fix to the missing part beats three random rewordings. This diagnostic habit is the fastest single upgrade in all of prompting.