Prompt Engineering
Write effective AI prompts that produce consistent, high-quality content across your workflows.
Prompt Structure
Effective prompts follow a consistent structure:
System Prompt
Defines the AI's role and behavior:
- Who/what the AI is (expert, writer, analyst)
- General capabilities and constraints
- Tone and style guidelines
- Output format expectations
User Prompt
Provides the specific task:
- Context and background
- Specific instructions
- Required elements
- Examples (if needed)
Writing System Prompts
System prompts set the foundation for consistent output.
Good System Prompt Example
Template
You are an expert web content writer specializing in {{industry}} websites.
Your writing style is:
- Professional yet approachable
- Clear and scannable
- Persuasive without being pushy
- SEO-optimized naturally
When creating content:
- Use active voice
- Keep paragraphs short (2-3 sentences)
- Include clear calls-to-action
- Write for the target audience: {{target_audience}}
Format requirements:
- Output valid HTML
- Use semantic heading structure (h1, h2, h3)
- Include appropriate meta description suggestions
Avoid Vague Instructions
Don't say "write good content" - specify what "good" means. Be explicit about quality criteria.
Writing User Prompts
User prompts provide the specific task with all necessary context.
User Prompt Structure
Template
Create a {{page_type}} page for {{company_name}}.
## Context
{{company_description}}
## Page Requirements
1. Hero section with compelling headline
2. Value proposition (3 key benefits)
3. Feature highlights with icons
4. Social proof section
5. Clear call-to-action
## Constraints
- Maximum 800 words
- Reading level: 8th grade
- Include keywords: {{keywords}}
## Previous Content
The homepage we created says: {{homepage_summary}}
Maintain consistency with this messaging.
Key Principles
- Be specific: "3 key benefits" not "some benefits"
- Provide context: Include relevant previous outputs
- Set constraints: Word count, format, reading level
- Use structure: Numbered lists and headers help AI parse
Output Formatting
Control output format for reliable parsing.
JSON Output
Template
Respond with a JSON object in this exact format:
{
"title": "Page title",
"meta_description": "160 characters max",
"sections": [
{
"type": "hero|features|cta",
"heading": "Section heading",
"content": "Section content HTML"
}
]
}
Output only the JSON object, no additional text.
HTML Output
Template
Output the content as valid HTML using these rules:
- Start with an h1 tag containing the page title
- Use h2 for section headings, h3 for subsections
- Wrap paragraphs in tags
- Use
or for lists
- Include class="cta-button" on call-to-action links
Do not include , , or tags.
Output only the content to be placed in the page body.
Common Issues and Solutions
| Issue | Cause | Solution |
|---|---|---|
| Inconsistent output format | Unclear format instructions | Provide explicit format examples |
| Generic content | Insufficient context | Add more specific details about business |
| Wrong tone | Vague style instructions | Give concrete examples of desired tone |
| Content too long/short | No length constraints | Specify word count or section limits |
| Hallucinated facts | AI filling gaps creatively | Provide all factual details explicitly |