Optimizing Prompts

Given this prompt: "Write about marketing"

Output:

Issues identified:

• No target audience specified
• No scope or angle defined
• No length, format, or tone guidance
• No success criteria

Optimized prompt:

Write a 600-800 word blog post about email marketing best practices for small e-commerce business owners with limited marketing experience.

Structure:
- Brief intro on why email marketing matters for e-commerce (2-3 sentences)
- 4-5 actionable tips, each with a short example
- Closing call-to-action encouraging list-building

Tone: Practical and encouraging, avoid jargon. Assume the reader has never run an email campaign before.

If a claim needs current data/statistics you're unsure about, flag it rather than inventing numbers.

Key improvements: Added audience, scope, structure, length, tone, and explicit permission to flag uncertainty instead of hallucinating.

Progress:

• Step 1: Analyze the original prompt for issues
• Step 2: Identify the core intent
• Step 3: Apply optimization principles
• Step 4: Consider advanced techniques if warranted
• Step 5: Present the optimized prompt with explanation
• Step 6: Iterate based on feedback

Step 1: Analyze the Original Prompt

Check for:

• Clarity issues — ambiguous terms, implicit assumptions, missing context
• Specificity gaps — no constraints, undefined success criteria, unclear scope/audience
• Structure problems — disorganized, no logical flow, no sections
• Format gaps — unspecified output format, length, tone, or examples
• Complexity — task too big for one prompt; needs chaining or step-by-step reasoning

Step 2: Identify the Core Intent

Determine: What is the user actually trying to accomplish? What defines success? Who's the audience? If unclear from the prompt, ask before optimizing rather than guessing.

Step 3: Apply Optimization Principles

• Be clear and direct — state requirements explicitly, remove vague language
• Add context — explain why requirements matter, include use case/background
• Add specificity — length, format, scope, audience, quality bar
• Structure the request — logical order, numbered points, separated sections (context / requirements / format)
• Include examples — especially for tricky formats, tone, or edge cases
• Allow uncertainty — explicitly permit "I don't know" to prevent hallucination

Step 4: Consider Advanced Techniques

Apply only when warranted by task complexity:

Load references/prompt-best-practices.md for detailed guidance and troubleshooting on these techniques.

Step 5: Present the Optimized Prompt

Use this exact output format:

**Analysis — issues identified:**
- [issue 1]
- [issue 2]

**Optimized Prompt:**
[full optimized prompt in a code block]

**Key improvements:**
- [improvement 1]
- [improvement 2]

**Usage tips (optional):**
[brief note on further customization]

Step 6: Iterate

Ask if the optimization meets their needs. Offer to adjust tone, length, specificity, or provide alternate formulations based on feedback.

Example 1: Vague → Specific Input: "Help me write an email to my team" Output: Optimized prompt specifying purpose (status update vs. request), recipient context (team size/role), key points to cover, desired tone (formal/casual), length limit, and a call-to-action requirement.

Example 2: Implicit context → Explicit Input: "Fix this code" (with code pasted, no other detail) Output: Optimized prompt asking model to state assumptions, specify the bug/error being fixed, the language/framework, whether to explain changes or just provide the fix, and to flag if multiple valid fixes exist rather than picking one silently.

Example 3: Single complex prompt → Chain Input: "Research a topic, write a report, and create a presentation" Output: Three sequential prompts — (1) research and summarize key findings with sources, (2) turn findings into a structured report with defined sections, (3) convert report into slide outline with one key point per slide — each consuming the prior step's output.

Consult references/examples.md for more before/after examples across domains.