Specify Your Output Format
The second principle is format specification. Even a perfect task description can fail if the output arrives in the wrong structure.
Why Format Matters
Without format instructions, the AI will guess what you want. It might return:
- A wall of text when you wanted bullet points
- A paragraph when you wanted JSON
- A casual essay when you wanted a structured table
Common Output Formats
| Format | When to Use | Prompt Phrase |
|---|---|---|
| Markdown | Documents, reports | "Format as markdown with headers" |
| JSON | API responses, data | "Return valid JSON matching this schema" |
| Table | Comparisons, schedules | "Present as a markdown table" |
| Bullet List | Summaries, action items | "List as bullet points" |
| Numbered Steps | Tutorials, processes | "Provide numbered step-by-step instructions" |
| Code | Programming tasks | "Write a Python function that..." |
| CSV | Data export | "Format as CSV with headers" |
Format Specification Examples
For a report:
Format the output as a markdown document with:
- An executive summary (3 sentences)
- Three H2 sections with supporting details
- A conclusion with action items as a bullet list
For structured data:
Return the result as a JSON object with these fields:
{
"title": "string",
"summary": "string (max 100 words)",
"tags": ["string array"],
"difficulty": "beginner | intermediate | advanced"
}
Pro Tips
- Show, do not tell — provide an example of the exact format you want
- Be explicit about nesting — if you want sub-bullets, say so
- Specify delimiters — for CSV, state the separator character
- Control length per section — "Each section should be 2–3 paragraphs"
Combining clear direction (Principle 1) with format specification (Principle 2) dramatically improves output quality.