Lesson 3
How to get the best results from any AI tool
Press ← → to navigate
The difference isn't the tool. It's how you ask.
"Make this better."
Result: Generic, unhelpful changes
"Rewrite this email to be more concise, keep a friendly tone, and end with a clear call to action."
Result: Exactly what you wanted
You don't need to be an expert on day one. Moving from vague to specific gets you 80% of the improvement.
The 4 pillars of a good prompt
Proven techniques that work
What goes wrong and how to avoid it
Working with AI output
Giving AI what it needs
Reusable prompts for common tasks
What do you want?
Be specific about the action.
What's the background?
Give relevant info the AI needs.
What are the boundaries?
Length, format, tone, what to avoid.
How should it respond?
Bullets, table, email, step-by-step.
Be specific about what you want the AI to do.
Help me with this email.
Write a professional email declining a meeting invitation. Keep the tone friendly and suggest rescheduling next week.
Context transforms results. Tell the AI what it needs to know.
Summarize this report.
Summarize this Q4 earnings report for a board presentation. Focus on revenue growth and highlight any metrics that changed more than 10% from Q3.
Set boundaries so the AI stays on track.
Constraints aren't limiting, they're focusing. The more specific your boundaries, the more useful the output.
Tell the AI how to structure its response.
Compare options side by side
Quick scannable points
Ready to send format
Ordered instructions
Decision support
Ready to use snippets
Help me write something about our product.
[Task] Write a product announcement email [Context] for our new project management tool launching next month, targeting small business owners [Constraints] Keep it under 150 words, enthusiastic but professional tone, no jargon [Format] Subject line + body with 3 bullet points of key features + CTA
If your prompt could mean 10 different things, it will. Be specific enough that two different people would interpret it the same way.
Before sending a prompt, ask yourself: Would someone else reading this produce the same result? If not, add more detail.
Proven techniques for better AI results
Assign the AI an expertise to shape its perspective and vocabulary.
Act as a [role] with expertise in [area]...
Break complex tasks into numbered steps for thorough results.
Analyze this business proposal: 1. First, summarize the key points 2. Then, identify the strengths 3. Next, list potential risks 4. Finally, give your recommendation with reasoning
This works because it forces the AI to think through each part instead of jumping to a conclusion.
Show the AI what you want by giving examples of input → output.
Convert these to professional subject lines: Casual: 'Hey, about that thing we discussed' Professional: 'Follow-up: Q4 Budget Discussion' Casual: 'Quick question about the project' Professional: 'Request for Clarification: Project Timeline' Casual: 'Running late to the meeting' Professional: ???
Ask the AI to show its reasoning. Best for complex analysis, math, and decisions.
"Should we expand to the European market?"
→ "Yes, Europe is a great market."
"Should we expand to the European market? Think through this step by step, considering costs, competition, regulations, and timeline."
→ Detailed analysis
Tell the AI what NOT to do. Boundaries prevent unwanted output.
Don't expect perfection on attempt one. The conversation IS the tool.
Iteration examples: "This is close but make it more casual" / "Good, now add a section about pricing" / "Shorten the intro and expand the conclusion"
Create fill-in-the-blank templates for tasks you do often.
Write a [type] email to [recipient]. Context: [background] Tone: [formal/casual/diplomatic] Key points to include: [points] Length: [short/medium/long] End with: [call to action]
Save templates like this and fill in the blanks each time. Consistency + speed.
| Pattern | Best For | Key Phrase |
|---|---|---|
| Role Prompting | Specialized expertise | "Act as a..." |
| Step-by-Step | Complex multi-part tasks | "First... then... finally..." |
| Few-Shot | Consistent formatting | "Here's an example..." |
| Chain of Thought | Analysis & decisions | "Think step by step..." |
| Constraints | Preventing unwanted output | "Do NOT..." |
| Iteration | Refining results | "This is close but..." |
| Templates | Repeated tasks | "[Fill in the blank]" |
The real power comes from combining multiple patterns.
[Role] Act as a senior marketing strategist. [Step-by-step] Analyze our Q4 campaign results: 1. Summarize performance across channels 2. Identify the top 3 performing campaigns 3. Explain why they worked 4. Recommend changes for Q1 [Constraints] Keep each section under 100 words. Use data from the attached report only. Don't suggest increasing the budget. [Format] Use headers and bullet points.
Common prompting mistakes and how to avoid them
"Make this better." "Help me with my project." "Write something about marketing."
The AI has to guess your intent, audience, format, and quality bar.
"Rewrite this paragraph to be more concise and professional. Target audience is C-level executives. Keep the same key message."
Specific = predictable results.
Dumping 10 tasks in one prompt. "Write the report AND make a presentation AND draft the email AND create a spreadsheet AND..."
One task at a time. Build up. "First, let's write the report. [After reviewing] Great, now let's turn the key findings into a 5-slide presentation outline."
AI handles focused tasks better than scattered mega-prompts. Break it up.
Copy-pasting AI output without reading it is the fastest way to embarrass yourself.
The 80/20 rule: AI gets you 80% of the way. Your job is to review, refine, and own that last 20%. This is where your expertise matters.
First result not perfect ≠ AI can't do it.
Most people quit after one try. The best results come from 2-3 rounds of conversation. Treat it like working with a colleague: give feedback, iterate.
AI doesn't say "I'm not sure." It presents everything with the same confidence. This includes made-up facts, wrong numbers, and plausible-sounding nonsense (hallucinations).
How to evaluate and refine AI output
Run through this checklist every time before using AI output.
Perfect is the enemy of done. If it's 90% right, fix the 10% and move on.
Tell the AI exactly what to change.
"This is close but change the tone to be more casual"
"You missed the part about the budget. Add that in"
"The intro is too long. Cut it in half"
"Give me 3 alternative versions of the opening paragraph"
AI remembers the conversation. Build on previous responses instead of starting over.
Each message adds to the context. You're sculpting the result through conversation.
Giving AI what it needs to deliver
Who you are, what the project is, who the audience is. "I'm a marketing manager preparing for a board meeting..."
Paste documents, data, examples, or previous work. "Here are last quarter's numbers: ..."
Your style, tone, format preferences. "I prefer bullet points over paragraphs, casual tone, and short sentences."
| Tool | How Context Works | Best For |
|---|---|---|
| ChatGPT | Paste context each conversation | Quick one-off questions |
| Claude.ai | Projects feature stores docs | Long document work, research |
| Claude Code | CLAUDE.md file + reads your codebase | Software projects, file work |
Write a welcome email for new customers.
Result: Generic, bland, could be for any company
Write a welcome email for new customers of CloudSync, a B2B file-sharing tool for law firms. Tone: professional but warm. Mention our 256-bit encryption and 24/7 support. Include a link to our onboarding guide.
Result: Specific, on-brand, ready to send
Write a [type of content] about [topic]. Audience: [who will read this] Tone: [formal / casual / diplomatic / urgent] Length: [word count or short/medium/long] Key points to include: - [point 1] - [point 2] - [point 3] Format: [email / report / blog post / summary] Additional notes: [anything else relevant]
Analyze [document/data/situation]. Goal: [what decision this supports] Focus on: [specific aspects to examine] Compare: [what to compare, if applicable] Output format: [summary / table / pros-cons / recommendation] Length: [brief / detailed] Important: [constraints, what to ignore, what to flag]
Help me brainstorm [topic/problem]. Context: [background on the situation] Constraints: [budget, time, resources, etc.] What's been tried: [previous attempts, if any] Generate: [number] ideas that are [realistic / creative / unconventional] For each idea, include: - Brief description - Why it could work - Potential challenges Prioritize by: [impact / feasibility / cost]
Prompting is a skill. Like any skill, it gets better with practice.
Lesson 3
What to do
The background
The boundaries
The shape
Practice makes progress