Prompting That Works

Get dramatically better results by communicating effectively with AI — the gap between what you ask and what you get.

INFO

• Time: ~35 minutes
• Difficulty: Beginner
• What you'll learn: The CRAFT framework, advanced features, and templates that consistently produce better results

---

This Page Covers

• Why Most People Under-Prompt - The gap between what you ask and what you want
• The CRAFT Framework - Context, Role, Ask specifically, Format, Test & iterate
• Advanced Features - Thinking mode, Research mode, and understanding Markdown output
• Prompts for Analysis - Templates for analyzing documents and data
• Prompts for Writing - Templates for different content types
• Prompts for Decision-Making - Structured comparison prompts
• Prompts for Learning - How to use AI as a teacher

---

Why Most People Under-Prompt

There's a gap between what you ask and what you actually want. When you type "write me an email," AI has to guess:

• Email to whom?
• About what?
• What tone?
• How long?
• What's the goal?

AI fills these gaps with generic assumptions. The result is generic output.

The fix is simple: Tell AI what you actually want. The CRAFT framework helps you do this systematically.

---

The CRAFT Framework

C - Context

Give AI the background it needs. Without context, AI makes assumptions that are often wrong.

Bad prompt:

Write an email about the project delay.

Good prompt:

I'm a project manager at a software consulting firm. We're 2 weeks behind on a client deliverable due to unexpected technical issues. The client is generally understanding but has mentioned budget concerns. Write an email...

Context to include:

• Who you are - Your role, expertise, relationship to the recipient
• The situation - What's happening, what led to this point
• The audience - Who will read this, what do they care about
• Constraints - Budget, time, technical limitations, company policies

---

R - Role

Assign AI a persona or expertise. This primes it to respond from that perspective.

Without role:

Explain blockchain to me.

(Generic, textbook explanation)

With role:

Act as a patient teacher explaining blockchain to a complete beginner. Use simple analogies and avoid technical jargon.

(Clear, accessible explanation)

Useful roles:

• "Act as an experienced [job title] who specializes in..."
• "You are a [expert type] helping a [audience type]..."
• "Respond as if you're [specific persona]..."

Why This Works

Role assignment activates relevant patterns from AI's training. A prompt starting with "Act as an experienced copywriter" draws on patterns from professional copywriting, not generic text.

---

A - Ask Specifically

Vague asks get vague responses. Be precise about what you want.

Vague:

Give me some marketing ideas.

Specific:

Give me 5 marketing taglines for a sustainable coffee brand. Each tagline should be under 8 words, emphasize environmental responsibility, and appeal to millennial professionals.

Things to specify:

• Quantity - "Give me 3 options" vs "give me ideas"
• Length - "In 2-3 sentences" or "a 500-word article"
• Style - "Professional but friendly" or "casual and conversational"
• Constraints - "Avoid industry jargon" or "must include a call to action"

---

F - Format

Tell AI how to structure the output. This saves you reformatting later.

Request formats like:

• "Format this as a table with columns for [X], [Y], and [Z]"
• "Use bullet points with bold headers"
• "Structure this as: Problem → Solution → Next Steps"
• "Present as a numbered list, ordered by priority"
• "Write in markdown with H2 headers for each section"

Example:

Compare these 3 project management tools. Format as a table with columns: Tool Name, Best For, Pricing, Main Limitation.

Tool Name	Best For	Pricing	Main Limitation
Asana	Teams needing task dependencies	Free - $25/user/mo	Complex for simple projects
Trello	Visual kanban workflows	Free - $17.50/user/mo	Limited reporting
Notion	All-in-one workspace	Free - $10/user/mo	Steeper learning curve

---

T - Test & Iterate

Your first response is rarely perfect. Iteration is expected and normal.

Refinement techniques:

• "Make it more concise"
• "Make it more formal/casual"
• "Add more detail about [specific aspect]"
• "Remove the section about [X]"
• "Rewrite this for [different audience]"

When to start over:

• The conversation has drifted far from your goal
• AI seems confused about what you want
• You've made many contradictory requests
• The context window is getting full (long conversation)

Pro Tip

If you get a great response, save the prompt. You can reuse it for similar tasks.

---

Advanced Features

Beyond basic prompting, modern AI tools have special modes that can dramatically improve results for specific tasks.

Thinking Mode (Extended Reasoning)

Both Claude and ChatGPT offer an "extended thinking" or "reasoning" mode. When enabled, the AI spends more time working through problems step-by-step before giving you a response.

In Claude: Look for "Extended thinking" in the model selector In ChatGPT: Select "GPT-5.2 Thinking" from the model selector (designed for complex reasoning)

When to use Thinking Mode:

• Complex analysis with multiple factors
• Math problems or calculations
• Multi-step reasoning tasks
• Code debugging
• Strategic planning with trade-offs

When NOT to use Thinking Mode:

• Simple questions with straightforward answers
• Creative writing (thinking can over-constrain it)
• When speed matters more than depth
• Casual conversation
• Tasks you've already done that worked well without it

Cost and Speed

Thinking mode uses more compute time, which means slower responses and (for paid tiers or API usage) higher costs. Use it deliberately for tasks that benefit from deeper reasoning.

---

Research Mode (Web Search)

ChatGPT and Claude can search the web in real-time to find current information. This is different from their "trained knowledge" which has a cutoff date.

In Claude: Look for the web search toggle or ask Claude to search In ChatGPT: Enable "Browse" or use a model with browsing capabilities

When to use Research Mode:

• Current events or recent news
• Today's prices, statistics, or data
• Fact-checking claims
• Finding specific sources or citations
• Anything where "as of today" matters

When NOT to use Research Mode:

• General knowledge questions (faster without search)
• Creative or hypothetical tasks
• Personal advice or analysis
• When you've already provided the information in your prompt

Always Verify

Even with search enabled, AI can misinterpret or summarize sources incorrectly. For important decisions, click through to the actual sources AI cites.

---

Understanding Markdown Output

When you copy text from ChatGPT or Claude, you might notice strange formatting:

**This text has asterisks**
# And hash symbols
- And dashes like this
[Links look like this](url)

This is Markdown - a simple formatting language. AI outputs are formatted in Markdown because it's clean and widely used.

Common Markdown symbols you'll see:

Symbol	Meaning
**text**	Bold
*text*	Italic
# Header	Large heading
- item	Bullet point
1. item	Numbered list
`code`	Code or technical term

When pasting AI output:

• If pasting into a Markdown-aware tool (Notion, GitHub, etc.) → formatting will render correctly
• If pasting into plain text (email, Word) → you may need to remove the formatting symbols
• If pasting into a rich text editor → some apps auto-convert Markdown

Quick Fix

If you want clean text without formatting, add to your prompt: "Respond in plain text without any formatting."

---

Prompt Templates

Copy and customize these templates for common tasks.

For Analysis

I need to analyze [DOCUMENT/DATA TYPE].

Context: [Brief background on what this is and why you're analyzing it]

Please:
1. Summarize the key points in [X] bullet points
2. Identify [specific things to look for - risks, opportunities, patterns]
3. Highlight anything unusual or concerning
4. Suggest [X] follow-up questions I should consider

Format the output with clear headers for each section.

Example filled in:

I need to analyze a vendor contract for our marketing agency.

Context: We're a 50-person tech company evaluating a 12-month contract with a new PR firm. This would be our largest marketing spend.

Please:
1. Summarize the key terms in 5 bullet points
2. Identify any unusual clauses or potential risks
3. Highlight anything that differs from standard agency contracts
4. Suggest 3 questions I should ask before signing

---

For Writing

Write a [CONTENT TYPE] about [TOPIC].

Context: [Who you are, the situation, any background]
Audience: [Who will read this, what they care about]
Tone: [Professional, casual, persuasive, informative, etc.]
Length: [Word count or structure guidance]

Key points to include:
- [Point 1]
- [Point 2]
- [Point 3]

Avoid: [Anything to exclude - jargon, certain topics, etc.]

Example filled in:

Write a LinkedIn post announcing our company's new sustainability initiative.

Context: I'm the CEO of a 200-person software company. We just achieved carbon neutrality.
Audience: Our employees, clients, and industry peers
Tone: Proud but humble, authentic, not preachy
Length: 150-200 words

Key points to include:
- The milestone we achieved
- One specific action we took
- Invitation for others to share their journey

Avoid: Greenwashing language, bragging, technical details

---

For Decision-Making

Help me decide between [OPTION A] and [OPTION B].

Context: [The situation and why this decision matters]

For each option, analyze:
- Pros
- Cons
- Risks
- Best suited for: [what scenario]

My priorities are: [List what matters most to you]

Format as a comparison table, then provide a recommendation based on my priorities.

Example filled in:

Help me decide between hiring a full-time content writer vs. using a freelance agency.

Context: We're a B2B SaaS startup that needs to produce 4 blog posts per month. Budget is $4-6K/month.

For each option, analyze:
- Pros
- Cons
- Risks
- Best suited for: what type of company

My priorities are: Consistent quality, someone who learns our industry, flexibility as needs change.

---

For Learning

Explain [CONCEPT] to me.

My background: [Your level of expertise in this area]
I learn best through: [Examples, analogies, step-by-step, visuals described, etc.]

Please:
1. Start with a simple explanation
2. Give a concrete example or analogy
3. Explain why this matters / when I'd use it
4. Point out common misconceptions
5. Suggest what I should learn next

If I ask follow-up questions, adjust your explanation based on what I'm struggling with.

Example filled in:

Explain how API rate limiting works.

My background: I'm a product manager, not a developer. I understand basic web concepts.
I learn best through: Real-world analogies and examples

Please:
1. Start with a simple explanation
2. Give a concrete example I'd encounter in my work
3. Explain why this matters when evaluating vendor APIs
4. Point out misconceptions non-technical people often have
5. Suggest what related concept I should understand next

---

Common Prompting Mistakes

1. Being Too Vague

Problem:

Help me with my presentation

Better:

Help me create 5 slides for a 10-minute presentation to our board about Q3 sales results. The main message is that we exceeded targets despite market challenges.

2. No Context

Problem:

Write a professional email

Better:

Write a professional email from a sales manager to a prospect who went silent after our demo last week. Goal is to re-engage without being pushy.

3. Asking for Too Much at Once

Problem:

Write me a complete marketing strategy with target audiences, messaging, channels, budget, timeline, and KPIs

Better: Break it into steps. Start with:

Help me define 3 target audience segments for [product]

Then build from there.

4. Not Iterating

Problem: Accepting the first response even if it's not quite right

Better:

Good start, but make it more concise

or

The tone is too formal, make it friendlier

5. Copying Without Review

Problem: Using AI output directly without reading it carefully

Better: Read everything. Edit for your voice. Verify facts. Make it yours.

---

Key Takeaways

• Use CRAFT: Context, Role, Ask specifically, Format, Test & iterate
• Specificity wins: The more precise your request, the better the output
• Iteration is normal: First response is rarely final - refine it
• Use Thinking mode for complex reasoning - Skip it for simple queries
• Use Research mode for current info - Skip it for general knowledge
• AI outputs Markdown - Be aware when copying to other applications
• Save good prompts: When something works, save it for reuse
• Stay in control: AI is a tool to accelerate your work, not replace your judgment