Advanced Prompting — Get More From Every Request
Going Beyond Basic Prompts
Module 1's prompting framework (role, task, context, format, constraints) gets you 80% of the way there. Now we add techniques that unlock the other 20% — the nuanced, complex, high-value work.
These techniques work everywhere Claude does. They sharpen your Chat answers, and they make Cowork noticeably better at real tasks — a clear goal, an example of "good," and explicit constraints are what turn a vague request into a finished deliverable.
Technique 1: Chain of Thought
Ask Claude to think through a problem step-by-step before giving you the final answer. This dramatically improves accuracy for analytical tasks.
How to use it:
"Before giving me your answer, think through this step by step."
or:
"Walk me through your reasoning as you answer this."
Example:
"I'm deciding whether to hire a full-time developer vs. use a freelancer for a 3-month project. The project pays $50k. Walk me through your reasoning step by step before giving me a recommendation."
When Claude externalizes its reasoning, you can catch where it goes wrong and correct it.
Technique 2: Provide Examples (Few-Shot Prompting)
Show Claude what "good" looks like by giving it examples before asking it to produce something.
How to use it:
"Here are 3 examples of the type of [thing] I want: [example 1], [example 2], [example 3]. Now write one in the same style for [new topic]."
Example:
"Here are 3 subject lines from emails I've sent that got great open rates:
- 'You're making this mistake every Monday morning'
- 'Quick question about your Q3 numbers'
- 'The one thing I'd change about your onboarding'
Write 5 more in this style for an email about a new pricing plan."
Claude is very good at pattern-matching from examples.
Technique 3: Persona Injection
Go beyond "act as an expert" — give Claude a fully fleshed-out persona with opinions, a communication style, and a point of view.
Example:
"You are Alex, a blunt and direct product manager who has shipped 12 consumer apps. You don't believe in padding feedback with compliments. Review my product roadmap and tell me exactly what's wrong with it."
vs.
"Act as a product manager and review my roadmap."
The first version produces much sharper, more useful feedback.
Technique 4: The Structured Output Request
When you need Claude's output to feed into something else (a spreadsheet, an email, a report), specify the structure precisely.
Example:
"Analyze these 5 customer reviews and output your analysis in this exact format:
Sentiment: [Positive/Negative/Mixed] Key themes: [bullet list] Actionable improvements: [numbered list, max 3] One-sentence summary: [single sentence]"
This is especially powerful when you're doing repetitive analysis — you can paste in new content and get consistently structured outputs.
Technique 5: Asking Claude to Critique Itself
After Claude gives you a response, ask it to find the weaknesses in its own answer.
"Now that you've given me that answer, what are the weaknesses or things you're uncertain about?"
"Play devil's advocate against your own recommendation. What are the strongest objections to what you just suggested?"
This surfaces blind spots and makes the final output more reliable.
Technique 6: Constraint Stacking
Layer specific constraints to narrow the output to exactly what you need.
Example:
"Write a tagline for my productivity app. Constraints:
- Must be under 8 words
- Cannot use the words 'simple', 'easy', or 'powerful'
- Must speak to the emotional benefit, not the feature
- Should work well as a spoken sentence (will be used in a podcast ad)"
The more specific the constraints, the less editing you'll need to do.
Technique 7: The Reframe
Sometimes Claude gives you a technically correct answer that misses the spirit of what you need. Use a reframe to redirect it.
"That's technically accurate, but I need you to approach this from a sales perspective — assume the reader is skeptical and needs to be convinced, not just informed."
"Rewrite this but imagine the reader has 30 seconds, not 3 minutes."
"Give me the same answer but aimed at a CEO, not a developer."
Technique 8: System Prompts via "Before We Start"
In a regular chat, you can set the stage before asking your actual question:
"Before we start: I want you to respond to everything in this conversation in plain language, avoid jargon, and keep responses concise. I also want you to push back on me if I say something that doesn't make sense. Now, [actual question]."
This functions like a mini system prompt for the conversation.
Combining Techniques
The real power comes from combining these:
"You are a sharp, experienced copywriter (persona) who doesn't waste words. Here are 2 examples of landing page headlines I like (few-shot): [examples]. Write 5 headlines for my SaaS product (task). Constraints: no buzzwords, no exclamation points, under 10 words each, focused on the outcome not the feature (constraints). After you write them, tell me which one you'd actually use and why (chain of thought)."
Summary
- Chain of thought: ask Claude to reason step-by-step before answering
- Few-shot: show examples of the output you want
- Persona injection: give Claude a specific, opinionated character
- Structured output: define exactly the format you need
- Self-critique: ask Claude to find the weaknesses in its own answer
- Constraint stacking: layer specific rules to narrow the output
- Reframe: redirect when the answer is technically right but misses the point