Conversation Styles & System Prompts
From the course Claude for Power Users: Master Anthropic's AI
One of Claude's greatest strengths is its ability to adapt its communication style. Learning to control this gives you a much better experience.
Setting the Tone
Claude responds well to explicit style instructions:
"Respond in a casual, friendly tone. Use short sentences. Include analogies."
"Be extremely concise. No introductions or conclusions. Just the facts."
"Write as if you're a senior engineer explaining to a junior colleague. Be encouraging but precise."
System Prompts (Custom Instructions)
On claude.ai, you can set a system prompt that applies to every message in a conversation. This is your most powerful tool for controlling Claude's behavior:
"You are a concise technical writer. Always:
- Use bullet points over paragraphs
- Include code examples for any technical concept
- Flag assumptions you're making
- Ask clarifying questions when the request is ambiguous"
Concise vs Detailed Modes
Claude tends to be thorough by default. Control verbosity explicitly:
- For concise answers: "Answer in 2-3 sentences max." or "Give me the TLDR."
- For detailed analysis: "Provide a comprehensive analysis. Be thorough."
- For structured output: "Return your answer as a numbered list with exactly 5 items."
The "Think" Prompt
Claude has a special ability when asked to think:
"Think carefully about this problem before answering. Show your reasoning in <thinking> tags, then provide your final answer."
This activates extended thinking mode, where Claude reasons through complex problems more carefully before responding.
Conversation Management
- Start new conversations for unrelated topics — context pollution reduces quality
- Use follow-ups for iterative refinement within a topic
- Reference earlier parts of the conversation explicitly: "In your first answer, you mentioned X. Expand on that."
- If Claude misunderstands, rephrase rather than saying "that's wrong" — it responds better to positive redirection
Key Takeaways
- •Explicitly set tone, length, and style in your prompts for better results
- •System prompts control Claude's behavior for the entire conversation
- •Use 'Think carefully' or <thinking> tags to activate extended reasoning
- •Start new conversations for unrelated topics to avoid context pollution
- •Rephrase rather than correct — Claude responds better to positive redirection
Progress is kept in this browser only.