Prompting 102: task, context, output, constraintsPrompting 102: task, context, output, constraints

A weak prompt usually hides the real assignment in your head.

You know the deadline, audience, tone, risk, and desired output. The AI does not.

The easiest fix is a four-part prompt:

Task
Context
Output
Constraints

Task

The task is the verb. It tells AI what kind of work to do.

Good task words:

• rewrite
• summarize
• compare
• explain
• translate
• draft
• critique
• extract
• classify
• plan

Weak:

Help me with this.

Better:

Summarize this meeting note into action items.

Better still:

Summarize this meeting note into action items for the project owner.

Context

Context is the information that changes the answer.

Ask yourself:

• Who is this for?
• What happened before this?
• What does the reader already know?
• What does the AI need to avoid?
• What would make the answer wrong?

Example:

Context:
- This is for a client who already approved the budget.
- The project is delayed by one week because the design assets arrived late.
- We want to stay accountable without blaming the client.

Good context is not everything you know. It is the information that helps the model choose correctly.

Output

Output tells the model what shape to produce.

Be concrete:

Output:
- Thai email
- 100-150 words
- subject line + body
- polite but direct tone

Without output instructions, the model chooses its own format. That might be a paragraph, a table, a list, or a lecture.

Constraints

Constraints are boundaries.

Examples:

Constraints:
- Do not mention price.
- Do not promise a delivery date.
- Keep the original meaning.
- Use plain Thai, not corporate jargon.
- If information is missing, ask before writing.

Constraints are where many prompts become professional. They prevent the model from being helpful in the wrong direction.

The four-part template

Copy this:

Task:
[what work should AI do?]

Context:
[who, situation, goal, background]

Output:
[format, language, length, tone]

Constraints:
[must keep, must avoid, ask if missing]

Example: rough to useful

Rough:

Make this sound better.

Useful:

Task:
Rewrite this message for a customer.

Context:
The customer asked why the invoice is higher than expected. The extra cost comes from two approved add-ons.

Output:
Thai message, 3 short paragraphs, polite and clear.

Constraints:
Do not sound defensive. Do not offer a discount. Mention that the add-ons were approved on May 3.

Text:
[paste message]

Practice

Take one real AI request you made recently. Rewrite it into the four-part template.

Then check:

• Does the task start with a clear verb?
• Did you include only useful context?
• Is the output shape visible?
• Did you add at least one boundary?

References

• Google Gemini Prompt Design Strategies recommends clear instructions, constraints, prefixes, and breaking complex prompts into components.
• OpenAI Prompt Engineering describes how instructions, examples, and context can be separated for clearer model behavior.
• Microsoft Foundry Prompt Engineering Techniques recommends clear syntax and task decomposition.