IMPORTANT TERMS
| Term | Definition |
|---|---|
| Prompt | The instruction or question you provide to Claude. |
| Output | The response generated by Claude. |
| Token | Smallest unit of text processed by Claude (roughly 4 characters). |
| Context Window | Claude’s memory capacity for the conversation (200K tokens). |
| Multimodality | Claude’s ability to process both text and images. |
| Reasoning | Claude’s process of working through complex problems step by step. |
| Constitutional AI | Anthropic’s approach to aligning Claude with human values. |
| Temperature | Parameter controlling randomness in Claude’s responses. |
| XML Tags | Special formatting that helps structure Claude’s responses. |
TONE SETTINGS
| Tone | Description |
|---|---|
| Academic | Scholarly, referenced, formal analysis. |
| Professional | Clear, concise business communication. |
| Expert | Detailed, technical, domain-specific knowledge. |
| Casual | Conversational, approachable, everyday language. |
| Creative | Imaginative, expressive, original thinking. |
| Empathetic | Understanding, supportive, emotionally aware. |
| Instructional | Clear, step-by-step guidance. |
| Balanced | Neutral, objective, considering multiple perspectives. |
| Enthusiastic | Energetic, positive, encouraging. |
| Thorough | Comprehensive, detailed, exhaustive. |
PROMPT STRUCTURES
Use these frameworks to structure your prompts for maximum clarity and precision:
| Framework | Description |
|---|---|
| CRISPE | Context, Role, Insight needed, Specifics, Process, Examples |
| ROSES | Role, Objective, Scenario, Expected Solution, Steps |
| STAR | Situation, Task, Action, Result |
| CAR | Context, Action, Result |
| GRADE | Goal, Request, Action, Detail, Examples |
| RACER | Role, Audience, Context, Execution, Requirements |
| CRAFT | Context, Requirements, Audience, Format, Tone |
| APE | Action, Purpose, Expectation |
| FORMAT | Function, Objective, Rules, Media, Audience, Tone |
| CARE | Context, Action, Result, Example |
EFFECTIVE COMMANDS
| Command | Description |
|---|---|
| Act as | Assign a specific role or expertise to Claude. |
| Break down | Analyze something step by step. |
| Summarize | Condense information to key points. |
| Compare and contrast | Identify similarities and differences. |
| Explain like I’m 5 | Simplify complex concepts. |
| Build on this | Continue developing previous ideas. |
| Create a table | Organize information in tabular format. |
| Use XML tags | Structure specific parts of the response. |
| Elaborate on | Provide more detailed explanations. |
| Critique this | Provide constructive feedback. |
PARAMETER INSTRUCTIONS
Temperature
Controls randomness in outputs. Lower values (e.g., 0.1) produce focused, consistent responses while higher values (e.g., 0.7) enable more creativity.
Example: “Please use a temperature of 0.2 for this factual analysis.”
Response Length
Guides how detailed or concise Claude should be.
Example: “Provide a comprehensive 1000-word analysis” or “Summarize this in 2-3 sentences.”
XML Tags
Requests specific formatting for structured outputs.
Example: “Organize your response using , , and tags.”
Format Control
Specifies the desired output structure.
Example: “Present this information as a markdown table with three columns.”
CLAUDE’S SPECIAL CAPABILITIES
| Capability | How to Use It |
|---|---|
| Multimodal Understanding | Upload images and ask Claude to analyze, describe, or extract information from them. |
| Extended Reasoning | Ask Claude to “think step by step” or activate extended thinking mode (Pro accounts). |
| Long Context Processing | Utilize Claude’s 200K token context window for analyzing long documents. |
| Code Generation | Request code with specific parameters, languages, and commenting styles. |
| Knowledge Through 2023 | For recent information beyond Claude’s knowledge, ask it to acknowledge limitations. |
| Structured Outputs | Request specific formats using XML tags like <table>, <list>, or custom tags. |
| Document Analysis | Upload PDFs, spreadsheets, or other documents for Claude to analyze. |
| Artifacts Generation | Ask Claude to create standalone text documents, code, HTML, SVG graphics, etc. |
MISTAKES TO AVOID
Vague Instructions
Avoid: “Tell me about history.”
Use: “Provide an overview of major technological developments during the Industrial Revolution (1760-1840), focusing on their economic impacts.”
Over-Constraining
Avoid: “Give me exactly 7 bullet points with precisely 10 words each about quantum computing.”
Use: “Explain the core principles of quantum computing in a concise bulleted list.”
Contradictory Requirements
Avoid: “Write a comprehensive but brief analysis.”
Use: “Write a focused 300-word analysis covering the three most important aspects.”
Leading Questions
Avoid: “Don’t you think climate change is exaggerated?”
Use: “What do scientific consensus reports indicate about the severity of climate change?”
Assuming Knowledge
Avoid: “Fix the errors in this code.”
Use: “This Python code should calculate Fibonacci numbers but contains errors. Please identify and fix them.”
ADVANCED TECHNIQUES
1. Few-Shot Learning
Provide examples of the desired input-output format before your actual request.
Example <span class="hljs-number">1</span>: <span class="hljs-string">Input:</span> Renewable energy investment <span class="hljs-string">Output:</span> Solar power, wind energy, hydroelectric, geothermal, biomass Example <span class="hljs-number">2</span>: <span class="hljs-string">Input:</span> Digital marketing channels <span class="hljs-string">Output:</span> Social media, email marketing, content marketing, SEO, PPC Now, please follow the same <span class="hljs-string">format:</span> <span class="hljs-string">Input:</span> Machine learning algorithms
2. Chain-of-Thought Prompting
Ask Claude to reason through problems step by step.
“Please think through this calculus problem step by step, explaining your reasoning at each point before giving the final answer.”
3. Self-Reflection Prompting
Ask Claude to evaluate its own responses.
“After providing your analysis, please review your answer and identify any potential weaknesses or areas where additional information would be helpful.”
4. Role Assignment Plus Audience Specification
Define both who Claude should act as and who the audience is.
“Act as an experienced data scientist explaining complex neural network concepts to marketing professionals with limited technical background.”
Created by @realpaulhebert http://www.RealPaulHebert.com
For the latest Claude capabilities, visit: anthropic.com/claude
