HeroPrompt
Back to docs
Prompts

Prompt Best Practices

Tips for getting the best results from HeroPrompt templates.

Updated 2026-02-15

Prompt Best Practices

Get better results from AI by following these proven strategies when using HeroPrompt templates.

1. Be Specific

AI models work best with clear, detailed instructions.

❌ Avoid vague requests:

text
"Make it better"
"Add some features"
"Improve the design"

✅ Be specific:

text
"Improve performance by reducing database queries — use caching"
"Add real-time validation with error messages below each form field"
"Update design to use Material Design 3 with rounded corners and elevated cards"

2. Provide Context

Give the AI background information to make informed decisions.

Essential context:

  • Target audience — Beginners, experts, general users?
  • Use case — Production code, prototype, learning exercise?
  • Constraints — Time, budget, technology limitations?
  • Environment — Languages, frameworks, platforms in use?

Example:

text
Context: Building a React dashboard for internal ops team (20 users).
Tech stack: React 18, TypeScript, TailwindCSS, React Query.
Priority: Fast development over perfect code. Must be mobile-responsive.

3. Use Examples

Show, don't just tell. Examples clarify expectations.

Instead of:

text
"Return the data in a good format"

Provide an example:

text
Return data in this format:
{
  "id": "user-123",
  "name": "Alice Smith",
  "created_at": "2024-01-15T10:30:00Z",
  "tags": ["admin", "verified"]
}

4. Specify Output Format

Tell the AI exactly how you want the response structured.

Format options:

  • Code only — No explanations, just implementation
  • Code + comments — Inline documentation
  • Step-by-step — Numbered instructions
  • Markdown — Headers, lists, code blocks
  • JSON — Structured data
  • Table — Comparison or feature matrix

Example:

text
Provide the response as:
1. Executive summary (3 sentences)
2. Implementation steps (numbered list)
3. Code example (TypeScript with comments)
4. Trade-offs (pros/cons table)

5. Set Constraints

Define what not to do.

Common constraints:

  • No external libraries — Use standard library only
  • Max lines — Keep functions under 50 lines
  • No premature optimization — Simple and readable first
  • Specific style — Follow PEP 8, ESLint Airbnb, etc.
  • Performance — Must handle 10k items without lag

Example:

text
Constraints:
- No React class components (hooks only)
- No external state management (use useState/useContext)
- Must work in browsers back to Chrome 90
- Total bundle size < 100KB

6. Ask for Reasoning

Request explanations to understand the AI's decisions.

Add to your prompt:

text
Explain your reasoning for:
- Why you chose this approach
- What alternatives you considered
- Any trade-offs made

This helps you evaluate the solution and learn best practices.

7. Iterate Incrementally

Start simple, then add complexity.

Good workflow:

  1. First prompt: Basic implementation
  2. Second prompt: "Add error handling"
  3. Third prompt: "Optimize for performance"
  4. Fourth prompt: "Add comprehensive tests"

Don't try to get everything perfect in one prompt.

8. Use Role-Play

Frame the AI as an expert in the domain.

Examples:

text
"You are a senior DevOps engineer with 10 years of Kubernetes experience..."
"You are a database performance specialist optimizing PostgreSQL queries..."
"You are a UX designer following Material Design 3 guidelines..."

This primes the AI to respond with appropriate expertise level.

9. Request Multiple Options

Ask for alternatives to compare.

text
Provide 3 approaches:
1. **Simple** — Easiest to implement, fewer features
2. **Balanced** — Good trade-off between complexity and features
3. **Advanced** — Full-featured, production-grade

For each, explain pros/cons and implementation complexity.

10. Check for Errors

Don't trust AI output blindly.

Always:

  • Run the code — Test before using in production
  • Review logic — Verify the approach makes sense
  • Check edge cases — Does it handle errors, empty inputs, etc.?
  • Validate — Use linters, type checkers, security scanners

11. Use the Right AI Model

Different models excel at different tasks.

TaskBest Model
Code generationClaude, GPT-4
Creative writingClaude, GPT-4
Data analysisGPT-4, Claude
Image generationDALL-E 3, Midjourney
Fast prototypingGPT-3.5 Turbo
Long contextClaude 2 (100k tokens)

Check the "AI Models" tag on each HeroPrompt for recommendations.

12. Leverage the Optimizer

Use the Prompt Optimizer to improve any prompt:

  • Enhance & Clarify — Make vague prompts specific
  • Structured Format — Organize complex requirements
  • Chain of Thought — Add step-by-step reasoning
  • Few-Shot Examples — Add input/output examples
  • Model-Specific — Tailor for Claude, GPT, or Gemini

Common Mistakes to Avoid

❌ Too vague — "Make it work"

✅ Specific — "Fix the authentication bug where users can't log in with special characters in passwords"

❌ No context — "Write a function"

✅ Context — "Write a Python function for a FastAPI backend that validates JWT tokens"

❌ Overloading — One massive prompt with 10 requirements

✅ Incremental — Break into smaller, focused prompts

❌ Assuming knowledge — Using jargon without explanation

✅ Clear — Define acronyms and domain-specific terms

Advanced: Prompt Chaining

For complex tasks, chain multiple prompts together:

text
Prompt 1: "Design the database schema for a blog platform"
↓ (use output)
Prompt 2: "Write SQLAlchemy models for this schema"
↓ (use output)
Prompt 3: "Create Alembic migrations for these models"
↓ (use output)
Prompt 4: "Generate API endpoints using FastAPI"

For automated prompt chaining, upgrade to PRO Kit for Skills.

Next Steps

Previous
Variables
Next
Optimizer Overview