Prompt Engineering for Claude

Same AI, wildly different results

Two people ask Claude to help with a marketing email. Person A types: "Write a marketing email." Person B types: "You're a senior email marketer at a DTC skincare brand. Write a re-engagement email for customers who haven't purchased in 90 days. Tone: warm but direct. Include a 15% discount code. Keep it under 150 words. Subject line should create urgency without being spammy."

Person A gets a generic email template. Person B gets a polished, ready-to-send email. Same AI. The only difference? The prompt.

The anatomy of a great prompt

Every effective Claude prompt has these building blocks:

You don't need all six every time. But the more you include, the better the output:

Technique 1: Be specific, not vague

The single biggest improvement you can make:

Technique 2: Give Claude a role

Starting with a role isn't just decoration — it fundamentally changes how Claude approaches the task:

You are a veteran startup CTO who has scaled three companies from
seed to Series C. You value pragmatic engineering decisions over
theoretical perfection. You've seen what works and what fails at scale.

Review my system architecture and tell me what will break first
when we go from 1,000 to 100,000 users.

The role sets:

• Expertise level — a CTO thinks differently from a junior developer
• Perspective — focuses on scaling, not academic correctness
• Communication style — direct, battle-tested advice

Technique 3: Use examples (few-shot prompting)

Showing Claude what you want is often more effective than describing it:

Convert these product descriptions into punchy one-liners for social media.

Input: "Our noise-cancelling headphones feature 40-hour battery life,
premium drivers, and adaptive EQ technology."
Output: "40 hours of silence. Your commute just became your sanctuary. 🎧"

Input: "Our ergonomic office chair has lumbar support, breathable mesh,
and 12 adjustable settings."
Output: "Your back called. It wants the chair with 12 ways to say 'thank you.' 🪑"

Now do these:
Input: "Our smart water bottle tracks hydration, glows to remind you
to drink, and keeps water cold for 24 hours."

By providing 2-3 examples, you've defined the tone, length, format, and creative style — all without explicitly stating any rules.

Technique 4: Chain of thought — "think step by step"

For complex reasoning tasks, asking Claude to show its work dramatically improves accuracy:

The step-by-step version forces Claude to reason through each part rather than jumping to a conclusion. This matters most for:

• Mathematical or logical problems
• Multi-factor decisions
• Tasks where the answer depends on intermediate analysis

Technique 5: Structured output

Tell Claude exactly how to format the response:

Analyze this customer feedback and return your analysis in this exact format:

**Sentiment**: [Positive/Negative/Mixed]
**Key themes** (bulleted list, max 5):
- theme 1
- theme 2

**Urgent issues** (numbered, with severity 1-5):
1. [Severity X] Issue description

**Recommended actions** (prioritized):
1. Action — Expected impact — Effort level

This works because Claude follows formatting instructions precisely. You get structured, consistent output every time.

Technique 6: Iterative refinement

Your first prompt rarely produces the perfect result. The skill is in the follow-ups:

Each follow-up makes the output closer to what you actually need. Don't try to get everything perfect in one prompt — iterate.

There Are No Dumb Questions

How long should a prompt be?

As long as it needs to be. A simple task might need one sentence. A complex task might need a full paragraph of context plus examples. There's no penalty for longer prompts — Claude handles them well. But don't add fluff — every word should carry information.

Should I be polite to Claude?

It doesn't affect output quality. "Please" and "thank you" are fine if that's your style, but they don't make Claude try harder. What matters is clarity, not politeness.

What if Claude's response is way off?

Don't start over — tell Claude what's wrong. "That's not what I meant. I need X instead of Y. Here's why..." Claude adjusts well to correction and often produces a much better result on the second try.

Technique 7: System prompts and custom instructions

When using Claude's API or Projects, you can set a system prompt — persistent instructions that shape every response:

System prompt:
You are a senior technical writer at a developer tools company.
Your writing style: clear, concise, no jargon unless necessary.
Always include code examples for technical concepts.
Format: Use H2 headers, short paragraphs (max 3 sentences), and
bullet points for lists.
Never say "simply" or "just" — nothing is simple to a beginner.

This is incredibly powerful because it:

• Eliminates repetitive instructions
• Ensures consistent output across conversations
• Sets the baseline so your individual prompts can focus on the task

Anti-patterns: what NOT to do