Skip to main content
In this lesson you’ll learn the structure behind prompts that reliably produce sharp, usable results. Twenty minutes earlier Maya’s first prompt produced a working email and a prospect reply within the hour. Buoyed by that win, she tried again for a different client: a yoga studio that wanted a punchy intro line for their new website. Maya typed, “Write a sales line for a yoga studio.” The result was generic and lifeless — something she could have copied from a search result. She spent 15 minutes trying to edit it into something usable, then gave up. Nothing about the tool had changed since her first success. So why did one prompt land and the other flop? The difference is structure. Prompts that get sharp answers usually include five pieces: a role, a task, context, constraints, and an output format. Miss too many of those, and the model has to guess.
A slide titled "A sharp prompt has five pieces" showing five connected icons labeled Role, Task, Context, Constraints, and Output. A warning below reads "Miss too many, and the model has to guess."

​
The five pieces, step by step

  1. Role — Tell the model who it should be. This narrows voice, assumptions, and knowledge.
  2. Task — The verb and the specific deliverable: write, summarize, translate, plan.
  3. Context — Audience, brand personality, examples to emulate or avoid, and any situational details.
  4. Constraints — Length limits, forbidden words, punctuation, or formality level.
  5. Output format — Paragraph, bullet list, table, or a single sentence.
Use the five-piece formula as a checklist when you construct prompts: Role, Task, Context, Constraints, Output. Not every prompt needs all five, but the more you include, the less post-editing you’ll do.

​
1) Role

Start with who the model should be. Naming a role biases the model toward a particular tone, vocabulary, and mental model. Example:
  • “You are a copywriter for wellness brands.”
A webpage or app screen titled "Tell DeepSeek who it is" explaining that a role narrows voice and knowledge. It shows an "Add a role" example: "You are a copywriter for wellness brands," plus a note about having a focused voice and the right knowledge.

​
2) Task

Be explicit about the verb and the deliverable. “Write a one-sentence intro line” is much sharper than “write a sales line.” Tips:
  • Use a clear verb (write, summarize, list, translate).
  • Specify scope (one sentence, three bullet points, a 200-word paragraph).

​
3) Context

Provide the situational details the model lacks: the target audience, brand voice, examples to use or avoid, and any relevant background. The yoga prompt failed because it lacked context, so the model defaulted to a generic answer.
A presentation slide titled "What DeepSeek doesn't know yet" showing a green lightbulb icon and three cards labeled "Who the audience is," "How the brand sounds," and "What tone to avoid." A red-bordered warning at the bottom reads "No context in → a generic line out."

​
4) Constraints

Constraints are the boundaries that keep the response usable: word limits, banned phrases, punctuation rules, or formality level. Example constraints:
  • Keep it under twenty words.
  • No exclamation marks.
  • Avoid industry jargon.
A slide titled "Set the boundaries" that reads "Constraints are the boundaries." It shows a scales icon and two checked constraints: "Keep it under twenty words" and "No exclamation marks."

​
5) Output format

Tell the model the shape you want back—one sentence, a bullet list, a table, or a code block. Explicit formatting reduces unnecessary variants and speeds up copy-ready results.
A UI mockup titled "Say the shape you want back" showing four format buttons — Paragraph, Bullet list, Table, and Code block — plus a highlighted guideline that reads "Give me one sentence and nothing else."

​
Quick reference table

​
A concrete rewrite and result

Maya rewrote the yoga prompt to include all five pieces:
  • Role: You are a copywriter for wellness brands.
  • Task: Write a one-sentence website intro.
  • Context: Audience: busy local adults; Brand: calm, modern; Highlight: 5 minutes from downtown.
  • Constraints: Under 20 words; no exclamation marks.
  • Output: One sentence only.
She sent it and the response was calm, specific, and matched the studio’s voice. The tool hadn’t changed — the prompt did — and what had been fifteen wasted minutes became ninety seconds of focused work.
A clean slide titled "All five pieces, one calm answer" showing a prompt breakdown (Role, Task, Context, Constraints, Output) for a copywriter. To the right is a sample one-sentence line for a yoga studio: "Move, breathe, and reset — modern yoga for a calmer week, five minutes from your door."

​
Why you still might see variance

Even well-structured prompts can produce different outputs across runs. Part of that variance comes from:
  • Model version and fine-tuning.
  • Sampling parameters like temperature and top_p.
  • Any system-level instructions or dataset differences.
If you need consistent, repeatable output, lower sampling randomness (e.g., lower temperature) and pin down the model version. For creative exploration, increase temperature or try multiple drafts.
Further reading: Now you have a practical formula: Role, Task, Context, Constraints, Output. Use it as a checklist to sharpen prompts, reduce editing time, and get copy-ready results faster.

Watch Video