Skip to main content

Module 1.2: The Art of Prompting — Communicating Effectively with Claude

Effective Prompting

The quality of Claude's output is directly proportional to the quality of your input. This lesson covers the prompting techniques that consistently produce excellent results.

Rule 1: Be Specific

Vague prompts get vague results. Compare:

Weak: "Write something about our product."

Strong: "Write a 300-word product description for our project management tool, aimed at CTOs at mid-size companies. Highlight the Gantt chart feature and SOC 2 compliance. Use a professional but approachable tone."

The strong prompt defines: length, topic, audience, key features, and tone. Claude does not have to guess any of these.

Rule 2: Provide Context

Claude performs dramatically better with context. Before asking it to write, analyze, or build something, provide:

  • Background: What is this for? Who will read/use it?
  • Examples: Show Claude what good output looks like
  • Constraints: What should it avoid? What format should it use?
  • Reference material: Paste in relevant documents, code, or data
Here is our brand voice guide:
[paste brand voice document]

Here is an example blog post in our style:
[paste example]

Now write a blog post about AI adoption for small businesses,
following the same style and format.

Rule 3: Structure Your Request

For complex tasks, break your prompt into sections:

## Task
Review the following API endpoint code for security vulnerabilities.

## Code
[paste code]

## Focus Areas
- SQL injection
- Authentication/authorization
- Input validation
- Error handling that leaks information

## Output Format
For each issue found:
1. Severity (critical/high/medium/low)
2. Location (file and line)
3. Description of the vulnerability
4. Recommended fix with code example

Rule 4: Iterate, Do Not Start Over

Claude's responses are rarely perfect on the first try. That is normal and expected. Instead of reprompting from scratch:

  • Give specific feedback: "The intro is good but too long. Cut it to 2 sentences."
  • Point to what works: "Keep the structure from sections 1-3, but rewrite section 4 to focus on ROI."
  • Add missing context: "I forgot to mention — our audience is non-technical. Remove the code examples and explain in plain English."

Rule 5: Set the Role

Telling Claude who it should be improves output quality:

You are a senior data analyst at a financial services company.
You specialize in customer churn analysis and have 10 years of
experience presenting findings to non-technical executives.

Analyze this customer data and prepare a summary for the CEO.

Common Prompting Mistakes

  • Being too polite: "Could you maybe try to write..." → Just state what you need clearly
  • Overloading one prompt: Asking for 5 different things in one message → Break into separate requests
  • Not providing examples: Expecting Claude to match your exact style without seeing it → Always include examples for style-sensitive work
  • Ignoring the output format: Getting prose when you wanted JSON → Explicitly specify the format

Key Takeaways

  • Specific prompts with context, constraints, and examples produce the best results.
  • Structure complex requests with clear sections and output format specifications.
  • Iterate on Claude's output with specific feedback instead of starting over.
  • Set roles and provide examples for style-sensitive work.

In Module 2, we will apply these techniques to real work — writing, coding, analysis, and decision-making.