The difference between a mediocre result and a genuinely useful one from Claude often comes down to how you write the prompt. These tips work whether you’re using Claude for writing, coding, research or analysis — and most of them take seconds to apply.
1. Be Specific About What You Want
Vague prompts produce vague results. The more specific you are about what you need, the better the output.
Instead of: “Write me something about cybersecurity.”
Try: “Write a 300-word plain English explanation of phishing attacks for employees who have no technical background. Use a real-world example.”
Specify the audience, length, tone, format and purpose whenever you can.
2. Give Claude a Role
Telling Claude to act as a specific type of expert helps it focus its responses appropriately.
Examples:
- “You are a senior IT support technician. A user has reported…”
- “Act as an experienced technical writer. Review this documentation and suggest improvements.”
- “You are a network engineer. Explain this error message to me.”
3. Ask It to Think Step by Step
For anything involving reasoning, analysis or problem-solving, asking Claude to work through the problem step by step consistently produces better results than asking for a direct answer.
Try adding: “Think through this step by step before giving me your answer.”
This is especially effective for debugging code, diagnosing technical problems, or working through complex decisions.
4. Paste the Full Context
Claude has a 200,000 token context window — that’s roughly 150,000 words. Don’t summarise what you’re working with if you can paste the whole thing. The more context Claude has, the more accurate and relevant its response.
Paste the full error message, the entire email thread, the whole section of documentation. Don’t trim it down and then wonder why the response misses the point.
5. Tell It the Format You Want
Claude will adapt to whatever format you specify. Be explicit about this.
- “Give me a numbered list of five options.”
- “Format this as a table with columns for Feature, Pros and Cons.”
- “Write this as bullet points, not paragraphs.”
- “Return only the code, no explanation.”
6. Ask for Options, Not Just One Answer
When you’re not sure what you need, ask Claude to give you multiple versions or approaches.
Example: “Give me three different ways to word this email — one formal, one friendly, one very brief.”
This is far more useful than accepting the first version and trying to edit it yourself.
7. Iterate — Don’t Start Again
Claude remembers the full conversation. You don’t need to start a new chat to refine a response. Use follow-up messages:
- “That’s good but make it shorter.”
- “The tone is too formal — rewrite it to sound more conversational.”
- “Add a section about security considerations.”
- “Now turn that into a step-by-step guide.”
8. Use Projects to Save Your Preferences
Claude’s Projects feature (available on Pro) lets you save a set of instructions that apply to every conversation in that project. Instead of repeating “always use UK spelling” or “I’m a Windows systems administrator” in every chat, set it once in the project instructions.
9. Tell Claude What You Don’t Want
Negative constraints are as useful as positive ones.
- “Don’t include an introduction — go straight to the content.”
- “Don’t add a conclusion or summary at the end.”
- “Avoid jargon — write for a non-technical reader.”
10. Ask Claude to Check Its Own Work
After getting a response, you can ask Claude to review it: “Is there anything in that response you’re not confident about?” or “Check that for factual accuracy.” This surfaces uncertainty that Claude might not have flagged unprompted.
