A useful prompt is a brief. It tells an AI assistant what finished work looks like, what information matters, which boundaries apply, and how the result should arrive.
This method works across general assistants because it is based on the task, not a secret command. It also makes weak output easier to diagnose: you can see whether the goal, context, constraint, or format was missing.
1. Goal: name an observable outcome
“Write about our workshop” names a topic. “Draft a 150-word invitation for first-time managers that explains the practical outcome and asks them to register by Friday” names a deliverable and action.
Use the outcome sentence:
Create [deliverable] for [audience] so they can [understand, decide, or act].
2. Context: include what changes the answer
Context is not every document you have. Include the audience’s prior knowledge, the situation, approved facts, source material, and examples that affect content, depth, or tone.
Label instructions and source material separately. If accuracy matters, state whether the assistant may use outside knowledge and what to do when the provided material is insufficient.
3. Constraints: create reviewable boundaries
Useful constraints include claims that must not be invented, facts that must be included, length, reading level, terminology, tone behaviors, privacy rules, and a clear fallback for uncertainty.
“Professional” is vague. “Direct, warm, and calm; use short sentences; avoid hype and exclamation marks” is visible in the output.
4. Format: design for the next step
Ask how the result will be used. A decision might need recommendation, evidence, risks, and unresolved questions. A spreadsheet workflow needs consistent fields. A meeting record needs decisions, actions, owners, dates, and unconfirmed items.
GOAL Create [specific deliverable] for [audience] so they can [action]. CONTEXT They already know: [...] Important situation: [...] Approved source material: <<< [...] >>> CONSTRAINTS - Include: [...] - Avoid: [...] - Voice and length: [...] - When information is missing: [label or ask] FORMAT Return [structure] with [required sections]. QUALITY Before finalizing, check against: [three criteria].
Use a brief–draft–review loop
Do not stop after the first output. Score the draft against criteria you wrote before seeing it. Identify the largest gap and request a scoped change:
The draft meets [criterion], but misses [criterion] because [evidence]. Revise only [scope] by [specific change]. Preserve [what already works].
Turn the prompt into a system
For repeated work, separate stable instructions from variable inputs. Stable elements include the process, source rules, boundaries, output schema, and quality rubric. Variable elements include the audience, objective, source material, deadline, and specific constraints.
Test at least three different cases. If the same failure repeats, improve the stable template rather than manually patching every answer.
Build and test the complete workflow in Prompt Like a Pro.