Most people get mediocre results from AI and blame the technology. The real problem? Their prompts are vague, context-free, and expect the AI to read their mind. A well-crafted prompt can transform a generic response into exactly what you need. This guide shows you the framework professionals use — with real examples you can copy today.
Why Most Prompts Fail
Before we fix your prompts, let's understand why they break. The most common mistakes:
- Vague instructions: "Write something about marketing" gives you generic fluff.
- No context: AI doesn't know your audience, goals, or constraints unless you tell it.
- Wrong model for the task: Using a creative model for analytical work, or vice versa.
- Expecting mind-reading: Assuming the AI understands your implicit requirements.
Here are four examples of bad prompts transformed into good ones:
Bad Prompt
Write about marketing.
Good Prompt
You are a senior marketing strategist. Write a 500-word LinkedIn post about B2B SaaS pricing strategies. Target founders at early-stage startups. Use a confident, data-driven tone. Focus on value-based pricing with 2-3 real-world examples.
Bad Prompt
Fix this code.
Good Prompt
You are a senior Python developer. Review this code for performance issues and bugs. The code processes CSV files with 1M+ rows. Suggest optimizations and explain each change. Output: refactored code with comments explaining improvements.
Bad Prompt
Help me with my business.
Good Prompt
I run a freelance graphic design business making $3K/month. I want to scale to $10K/month in 6 months. Current clients: 3 retainer clients at $1K each. Challenge: finding time for outreach while delivering work. Give me a prioritized action plan with specific tactics.
Bad Prompt
Summarize this article.
Good Prompt
Summarize this article for a busy executive. Format: 3 bullet points (max 20 words each) + 1 key insight + 1 recommended action. Focus on business implications, not technical details.
The CRISP Framework
Every effective prompt includes five elements. Remember them with CRISP:
C — Context
Set the scene. Who are you? What's the situation? What background information matters?
Without Context
Write an email to schedule a meeting.
With Context
I'm a product manager at a fintech startup. I need to schedule a 30-minute call with the VP of Engineering to discuss API integration timelines for our Q3 launch. We've exchanged emails before but never met. She's based in London (I'm in New York).
R — Role
Assign the AI a specific persona. This shapes the tone, expertise level, and perspective.
Without Role
Explain machine learning.
With Role
You are a computer science professor teaching an introductory ML course to business students. Explain machine learning using analogies from everyday life. Avoid jargon. Your goal is to build intuition, not technical depth.
I — Instructions
Be specific about what you want. Include format, length, style, and any constraints.
Vague Instructions
Create a social media post about our product.
Specific Instructions
Create a Twitter thread (5 tweets) announcing our new AI writing assistant. Include: hook in tweet 1, key features in tweets 2-3, social proof in tweet 4, CTA in tweet 5. Use emojis sparingly. Max 280 characters per tweet. Tone: excited but professional.
S — Style
Define the voice and tone. Should it be formal? Conversational? Technical? Persuasive?
No Style Guidance
Write a landing page for our consulting service.
With Style Guidance
Write a landing page for our executive coaching service. Tone: authoritative but warm, like a trusted mentor. Avoid hype and exclamation points. Use short paragraphs and conversational language. Target: C-suite executives at Fortune 500 companies.
P — Parameters
Set boundaries: word count, format, what to include or avoid, output structure.
No Parameters
Give me ideas for blog posts.
With Parameters
Generate 10 blog post ideas for a B2B SaaS company. Each idea should include: title (under 60 characters), target keyword, and 1-sentence premise. Focus on topics that rank for high-intent keywords. Avoid: AI-related topics (too saturated). Format: markdown table.
5 Prompt Templates You Can Use Today
Copy these templates and fill in the brackets:
1. Research Assistant
You are a research analyst specializing in [INDUSTRY].
Research [TOPIC] and provide:
1. Executive summary (3 sentences)
2. Key findings (5 bullet points)
3. Data sources with links
4. Contrarian viewpoints
5. Recommended next steps
Focus on information from the past 12 months. Prioritize primary sources over news articles.
2. Code Reviewer
You are a senior [LANGUAGE] developer with 15 years of experience.
Review this code for:
- Security vulnerabilities
- Performance bottlenecks
- Best practice violations
- Readability issues
For each issue found, provide:
- Line number
- Severity (Critical/High/Medium/Low)
- Explanation
- Suggested fix with code
Output format: Markdown table followed by refactored code.
3. Content Writer
You are a content writer for [PUBLICATION TYPE].
Write a [WORD COUNT]-word article about [TOPIC].
Audience: [DEMOGRAPHICS]
Tone: [TONE]
Goal: [DESIRED OUTCOME]
Requirements:
- Include [NUMBER] subheadings
- Add a compelling hook in the first paragraph
- End with a clear CTA
- Use short sentences (avg 15 words)
- Include 2-3 data points or statistics
4. Data Analyst
You are a data analyst. Analyze this dataset and provide:
1. Key insights (top 5)
2. Trends and patterns
3. Anomalies or outliers
4. Recommendations based on findings
Format insights as: "[Insight]" — [Supporting data] — [Business implication]
If you need clarification on the data, ask before proceeding.
5. Creative Brainstorm
You are a creative director at a top advertising agency.
Brainstorm [NUMBER] ideas for [PROJECT].
Constraints:
- Budget: [BUDGET]
- Timeline: [TIMELINE]
- Must appeal to: [TARGET AUDIENCE]
For each idea, provide:
- Concept name
- 1-sentence description
- Why it works
- Potential risk
Rank ideas by feasibility (1 = easiest to execute).
Advanced Techniques
Chain-of-Thought Prompting
For complex reasoning tasks, ask the AI to think step-by-step:
Chain-of-Thought Example
Solve this problem step by step. Show your reasoning at each stage.
Problem: A SaaS company has 10,000 users. 20% are on the free tier, 60% on basic ($29/mo), and 20% on pro ($99/mo). Churn rate is 5%/month. Customer acquisition cost is $150. What's the monthly revenue? How many new users needed monthly to grow 10%?
Few-Shot Prompting
Give the AI examples of the output you want:
Few-Shot Example
Convert these product features into benefit statements:
Feature: 256-bit encryption
Benefit: Your data is protected by bank-level security
Feature: Real-time collaboration
Benefit: Your team works together without version conflicts
Feature: AI-powered suggestions
Benefit: [Generate this]
Feature: Offline mode
Benefit: [Generate this]
System Prompts
For ChatGPT and Claude, you can set persistent instructions:
System Prompt Example
You are an expert copywriter. Always:
- Use active voice
- Write at an 8th-grade reading level
- Include one unexpected insight per response
- End with a question that prompts deeper thinking
- Never use the phrase "in conclusion"
Temperature Tuning
Control creativity vs. consistency via API:
- Temperature 0.0-0.3: Factual, consistent outputs. Best for code, analysis, factual tasks.
- Temperature 0.4-0.7: Balanced. Good for general writing, brainstorming.
- Temperature 0.8-1.0: Creative, varied outputs. Best for creative writing, ideation.
Model-Specific Tips
ChatGPT (GPT-4 / GPT-4o)
- Responds best to structured instructions with clear formatting requirements
- Excellent at following multi-step instructions
- Use numbered lists and explicit output formats
- Works well with the "act as" role-playing approach
- Best for: General tasks, multimodal work, creative projects
Claude
- Excels with nuance and long context (up to 200K tokens)
- Better at following complex, detailed instructions
- More thoughtful and measured responses
- Use XML tags to structure your prompt:
<context>...</context> - Best for: Long documents, nuanced analysis, coding, research
DeepSeek
- Best for code and reasoning tasks
- Excellent at mathematical and logical problems
- Strong performance on coding benchmarks
- Use chain-of-thought for complex reasoning
- Best for: Coding, math, technical analysis, Chinese-language tasks
Quick Reference Card
Before You Hit Send, Check:
- Context: Did I provide background information and constraints?
- Role: Did I assign a specific persona or expertise level?
- Instructions: Are my requirements specific and measurable?
- Style: Did I define the tone, voice, and audience?
- Parameters: Did I set word count, format, and boundaries?
200+ Ready-to-Use Prompts
Get our complete prompt library — tested templates for writing, coding, research, marketing, and more. Copy, paste, customize.
ChatGPT Ultimate Cheatsheet $12 →