Talking to AI, Prompt Engineering from Scratch

Why This Module Matters Most

If you take only one thing away from this entire course, let it be this: the way you talk to AI determines what you get back.

That sounds obvious, but most people miss it. They type a quick sentence, get a mediocre result, and conclude that AI is overhyped. They tried it once, it didn't impress them, and they moved on.

Here is what actually happened: they gave AI a bad instruction and blamed the AI for the bad result.

Think of it like hiring an assistant. If you tell a new hire "write me an email," what do you get? Probably something generic, maybe not even the right kind of email. You'd never do that with a person. You'd say, "I need to decline the job offer from Acme Corp. Keep it professional but warm. I want to leave the door open for future opportunities. Keep it under 150 words. Oh, and address it to Sarah, the hiring manager."

Same assistant. Different instructions. Different result.

AI works the same way. It is remarkably capable when you give it clear direction, and remarkably mediocre when you don't. The gap between "AI is meh" and "AI is amazing" is almost always the prompt.

The good news: this is a skill. It's not a talent you're born with or a secret code only engineers know. It's a learnable, repeatable way of communicating. By the end of this module, you'll have a framework and a set of go-to prompts that will change what you get out of every AI conversation from here on out.

The Five Rules of Good Prompts

Before we get into frameworks and templates, let's establish the principles. These five rules apply to every prompt you'll ever write.

Rule 1: Be Specific

Vague prompts produce vague results. Specific prompts produce specific results.

Vague: "Write an email."

Specific: "Write a professional email declining a job offer from Acme Corp, keeping the door open for future opportunities, under 150 words."

The second prompt tells AI exactly what the email needs to accomplish, who it's for, what tone to strike, and how long it should be. There's almost no room for misunderstanding.

This doesn't mean every prompt needs to be a paragraph. It means you should include the details that matter for the result you want. If length matters, say so. If tone matters, say so. If there's a specific point that must be included, say so.

Ask yourself: "If a smart but brand-new coworker read this, would they know exactly what I want?" If not, add more detail.

Rule 2: Give Context

AI doesn't know who you are. It doesn't know your situation, your audience, or your goals unless you tell it.

Context is the background information that helps AI give you a relevant answer instead of a generic one. Who are you? Who is this for? What's the situation? What are you trying to achieve?

Without context: "Write a cover letter for a marketing job."

With context: "I'm a career changer transitioning from teaching to marketing. I have 8 years of experience creating lesson plans, managing classroom projects, and communicating with parents. Write a cover letter for a content marketing role at a mid-size education technology company that highlights my transferable skills."

Same request, completely different result. The first one could be for anyone. The second one is for you.

Context is the single biggest factor in whether AI gives you something useful or something generic. When in doubt, add more context.

Rule 3: Show, Don't Just Tell

One of the fastest ways to get better results from AI is to show it what "good" looks like. Give it an example of the output you're aiming for.

This works because AI is excellent at pattern matching. When you provide an example, you're giving it a pattern to follow: the structure, the level of detail, the voice, the format.

Without an example: "Write a product description for my candle business."

With an example: "Write a product description for my candle business. Here's the style I'm going for:

'Our lavender chamomile candle fills your room with the calm of a summer evening. Hand-poured with 100% soy wax. 45-hour burn time. No synthetic fragrances. Just real lavender, real chamomile, real quiet.'

Now write something similar for our cedar and vanilla candle."

The example does more work than any instruction could. It shows the exact voice, length, and level of detail you want. AI will match that pattern closely.

You don't always need an example, but when the style or format matters, showing is faster and more reliable than describing.

Rule 4: Iterate

Your first prompt is rarely your best prompt. This is normal.

Most people type something, get a result, and either accept it or give up. The real skill is in the refinement. You read the output, figure out what's off, and adjust your prompt.

Maybe the tone is wrong. Maybe it's too long. Maybe it missed a key point. Instead of starting over, you follow up: "Make it more concise," or "Add a sentence about our refund policy," or "This is too formal. Rewrite it like I'm talking to a friend."

Iteration is where the magic happens. Each round of back-and-forth sharpens the result. Think of it like editing a draft. The first pass is never the final version.

A practical tip: before you start, give yourself permission for three rounds. Prompt, review, refine. That's the process. If the first response was perfect, great. But expecting perfection on the first try is like expecting a first draft to be publication-ready. It happens sometimes, but it's not the norm.

Rule 5: Break It Down

Complex tasks produce messy results when you try to do them all at once. The solution is to break them into smaller, simpler prompts.

Say you want to plan a week-long trip to Portugal. You could write one massive prompt: "Plan my 7-day trip to Portugal including flights, hotels, daily itinerary, restaurants, and budget." That might work, but you'll probably get a surface-level answer that covers everything and excels at nothing.

Instead, break it into steps:

"I'm planning a 7-day trip to Portugal in September. What are the best regions to visit for a first-timer who loves food, history, and the ocean?"
"Based on those regions, suggest a day-by-day route that minimizes travel time."
"For day 3 in Porto, create a detailed itinerary including breakfast, morning activity, lunch, afternoon activity, and dinner. I prefer local spots over tourist traps."
"What should I budget for this trip? Mid-range spending. Two people."

Each prompt builds on the previous answer. Each one is specific enough to get a useful result. And you stay in control the whole time because you can redirect at any step.

This approach takes a few more messages, but the result is dramatically better. And it's actually faster than trying to fix one giant, messy response.

The CRAFT Framework

The five rules are principles. The CRAFT framework is a practical tool you can use every time you sit down to write a prompt.

CRAFT gives you a structure so you don't forget the important parts. You won't always need every letter, and you won't always use them in order. But when you're stuck or getting bad results, run through CRAFT and see what's missing.

Here's what each letter stands for:

C -- Context

Tell AI the situation. Who are you? What role should AI play? What's the background?

Example: "You are an experienced career coach. I am a mid-level software engineer looking to transition into product management."

R -- Request

State exactly what you want AI to do. Be direct.

Example: "Write a LinkedIn summary for my profile that positions me for product management roles."

A -- Audience

Who will read or use this? A board of directors reads differently than a group of college students.

Example: "This is for recruiters and hiring managers at tech companies who will skim my profile in about 10 seconds."

F -- Format

How should the output be structured? Bullet list? Email? Table? Paragraph? Specific length?

Example: "Format as a single paragraph, no more than 130 words."

T -- Tone

What voice or style should it have?

Example: "Write in a confident but approachable tone. Professional, not stiff."

CRAFT in Action: Three Worked Examples

Example 1: The Professional Email

You need to tell a client that a project will be delayed.

C: "You are a professional project manager. I am a freelance designer working with a small business client. The client is not very technical and tends to worry easily."
R: "Write an email informing the client that their website redesign will be delayed by one week."
A: "This is for the client, a small business owner who values clear communication and reassurance."
F: "Format as an email, under 200 words. Start with the delay, then the reason, then the new timeline, then next steps."
T: "Tone: warm, transparent, and confident. Not apologetic -- we're managing the project proactively."

Result: A clear, reassuring email that delivers bad news without causing panic. The client knows exactly what's happening, why, and what comes next. No jargon, no over-apologizing, no ambiguity.

Compare that to: "Write an email saying the project is delayed." You'd get something generic, probably either too apologetic or too terse, and the client would be left anxious.

Example 2: The Learning Prompt

You're trying to understand a concept that keeps coming up at work.

C: "You are a patient teacher who specializes in explaining technical concepts to non-technical people. I am a marketing manager with no coding background."
R: "Explain what an API is and why my team keeps talking about it."
A: "This is for me -- I need to understand this well enough to have an informed conversation with my engineering team."
F: "Format: a short paragraph explaining what it is, then an everyday analogy, then 2-3 sentences on why it matters in practice. Under 200 words total."
T: "Tone: friendly and clear. No jargon. If you use a technical term, define it immediately."

Result: A clean explanation that uses an analogy (like a restaurant menu -- the API is the menu that tells you what you can order from the kitchen), connects it to your real situation, and gives you the confidence to participate in the conversation.

Compare that to: "What is an API?" You'd get a Wikipedia-style definition that's technically accurate and completely useless in a meeting.

Example 3: The Decision Helper

You're deciding between two job offers and can't figure out which one to take.

C: "You are a career advisor. I am a graphic designer with 5 years of experience, currently earning $72,000. I value work-life balance and creative freedom more than salary alone."
R: "Help me compare these two job offers and decide which one is better for me."
A: "This is for me to make a personal decision. I want clarity, not a recommendation."
F: "Format as a comparison table with these columns: Factor, Offer A, Offer B, My Priority (high/medium/low). Then give me 3 questions to ask myself that might help me decide."
T: "Tone: thoughtful and objective. Don't lean toward either option. Help me think, don't tell me what to do."

Result: A structured comparison that surfaces the tradeoffs, plus three probing questions that help you clarify what you actually want. You make the decision, but with much better information.

Compare that to: "Which job should I take?" You'd get a generic answer that doesn't account for what matters to you.

Notice something about these examples: none of them are long. The CRAFT elements add maybe two or three sentences to your prompt. You're not writing a novel. You're just filling in the details that make the difference between a generic result and a useful one.

Also notice that you don't always need every letter. If format doesn't matter, skip it. If the audience is obvious, skip it. CRAFT is a checklist, not a rigid formula. Use what helps.

The Most Useful Prompts for Everyday Life

You don't need to reinvent the wheel every time you talk to AI. Certain prompt patterns come up again and again. Here are six of the most useful ones, each with a template you can adapt, a concrete example, and notes on what to watch for.

1. Summarize This

Template: "Summarize this [article/document/email thread] in [number] sentences, focusing on [what matters to you]."

Example: "Summarize this article in 3 sentences, focusing on the implications for small business owners."

What to watch for: AI summaries sometimes smooth over nuance or miss a critical detail. Always skim the original if the stakes are high. Also, specifying what you care about ("focusing on cost implications" or "focusing on the timeline") gives you a much more useful summary than a generic one.

2. Explain This Simply

Template: "Explain [concept] to me like I'm [your level]. Use an everyday analogy."

Example: "Explain blockchain to me like I'm a smart person who has never worked in tech. Use an everyday analogy."

What to watch for: The analogy is the key. Without it, you get a simplified textbook definition. With it, you get something you can actually picture and remember. Also, be honest about your level. Saying "like I'm five" when you actually have some background wastes the detail you could get. Saying "like I'm an expert" when you're a beginner leaves you confused.

3. Help Me Draft an Email

Template: "Draft a [tone] email to [recipient] about [topic]. Key points to include: [list them]. Keep it under [word count]."

Example: "Draft a friendly but professional email to my landlord about renewing my lease. Key points: I want to renew, I'd like to negotiate the rent down by $50/month based on market rates, and I've been a reliable tenant for 3 years. Keep it under 150 words."

What to watch for: AI-written emails tend to be either too formal or too flowery. Always read the draft in your own voice before sending. Edit it until it sounds like you. Also, listing your key points explicitly prevents AI from inventing arguments you don't want to make.

4. Compare These Two Options

Template: "Compare [Option A] and [Option B] for someone who [your situation/priorities]. Focus on [what matters to you]."

Example: "Compare a Honda CR-V and a Toyota RAV4 for a family of four who drives mostly in the city, values reliability, and plans to keep the car for 10 years. Focus on long-term cost of ownership and safety ratings."

What to watch for: AI can sometimes weight factors differently than you would. It might emphasize price when you care more about reliability. That's why you specify your priorities. Also, AI's training data has a cutoff, so for anything time-sensitive (like current pricing or recent model changes), double-check with a current source.

5. Create a Step-by-Step Plan

Template: "Create a step-by-step plan to [goal]. I am starting from [current situation]. I have [constraints/resources]."

Example: "Create a step-by-step plan to start a small Etsy shop selling handmade candles. I am starting from scratch -- I've never sold anything online. I have about $500 to invest and 10 hours a week to work on this."

What to watch for: AI tends to skip steps that seem obvious. If you say "I'm starting from scratch," be specific about what that means. "Starting from scratch" to AI might still assume you have certain tools or knowledge. Also, ask AI to number the steps and estimate time for each. That helps you see if the plan fits your actual schedule.

6. What Are the Pros and Cons

Template: "List the pros and cons of [decision/option]. For context, I am [your situation]. Weight [your priority] more heavily."

Example: "List the pros and cons of working remotely full-time vs. hybrid (3 days in office). For context, I live 45 minutes from the office, I'm an introvert, and I'm worried about missing out on promotions by not being visible."

What to watch for: AI will give you a balanced list by default. If you have a strong preference or concern, name it. Otherwise, the pros and cons list will be generically balanced and not very helpful for your specific situation. Also, ask AI to go deeper on the items that matter most: "Tell me more about the promotion risk of being fully remote."

Advanced Moves (Still Beginner-Friendly)

These aren't advanced in the sense of being difficult. They're advanced in the sense that most people don't think to try them. Each one dramatically improves your results with minimal extra effort.

Ask AI to Ask You Questions First

This is the single most underrated prompting technique.

Instead of writing a long, detailed prompt and hoping you covered everything, try this:

"Ask me 3 questions before you write this email."

Now AI will ask you things you might not have thought to include. What's your relationship with the recipient? How urgent is this? Is there anything you definitely don't want to mention? You answer those questions, and now AI has the context it needed all along.

This works for almost anything. Writing a resume? "Ask me 5 questions about my experience before you draft it." Planning a trip? "Ask me what I care about most before you suggest an itinerary." Creating a presentation? "Ask me about my audience before you outline it."

You're basically turning AI into an interviewer. It asks, you answer, then it delivers something much better than if you'd just given it a one-line instruction.

This is also great when you're not sure what information AI needs. Let it tell you.

Chain Prompts

Instead of asking for the final product in one go, build it in stages.

Let's say you need to write a blog post about time management for freelancers. Instead of one prompt, try this sequence:

Prompt 1: "I want to write a blog post about time management for freelancers. Brainstorm 5 angles I could take. Make them specific, not generic."

Prompt 2: "I like angle 3. Now outline the post: introduction, 4 main sections, and a conclusion. Brief notes for each section."

Prompt 3: "Write the introduction based on the outline. Make it engaging. Under 200 words."

Prompt 4: "Now write section 1. Include a short example or story."

And so on.

Why does this work better? Because at each step, you're making a decision. You pick the angle. You review the outline before it gets fleshed out. You steer the direction. The final result is something you actually shaped, not something AI guessed at.

It takes a few more messages, but the quality difference is enormous. And you spend less time rewriting because you caught problems early, before they propagated through the whole piece.

Use AI to Improve Its Own Output

After you get a response, ask AI to critique it.

"Review this and suggest 3 improvements."

Or more specifically: "This is good but feels too generic. What would make it more specific and memorable? Suggest 3 changes."

Or: "Rate this draft from 1-10 for clarity, tone, and completeness. Explain your rating and suggest improvements."

This is like having a built-in editor. AI is surprisingly good at identifying weaknesses in its own output, probably because evaluating writing is easier than creating it. And because you're asking for suggestions, not a rewrite, you stay in control of which changes you accept.

You can also ask it to rewrite with a specific improvement: "Take this and make it 30% shorter while keeping all the key points." Or: "Rewrite this for someone who knows nothing about the topic."

The point is: the conversation doesn't end with the first response. That's just the starting point.

Common Mistakes

Knowing what to do is half the battle. Knowing what not to do is the other half.

Being Too Vague

This is the number one mistake, and it causes the number one complaint: "AI gave me something generic."

Yes, it did. Because you gave it a generic instruction.

"Help me with my resume" is not a prompt. It's a topic. A prompt would be: "I'm a project manager with 6 years of experience in healthcare IT. Rewrite my resume summary to highlight my track record of delivering projects on time and under budget. Under 50 words. Professional tone."

The fix is simple: before you hit enter, read your prompt back. Would a competent stranger know exactly what to produce? If not, add details.

Asking Once and Accepting the First Response

The first response is a draft. Treat it like one.

Many people type a prompt, get a response, and use it as-is. Sometimes that works. Often it doesn't. The real value comes from the back-and-forth. Ask for a revision. Point out what's wrong. Ask for a different angle. Push for something better.

This isn't about being a perfectionist. It's about recognizing that one round of feedback can turn a B- result into an A result with very little extra effort.

Not Giving AI Enough Context

"I need a marketing plan" is a prompt that could apply to a Fortune 500 company or a dog walking business. AI will give you something, but it won't be right for your situation.

Context answers the questions AI would ask if it could: Who are you? What are you trying to achieve? What are your constraints? Who is this for? What does success look like?

You don't need to write a biography. Two or three sentences of context can change everything. "I'm a solo real estate agent in a competitive suburban market. I have a $500/month marketing budget. I want to attract first-time homebuyers."

Now AI has something to work with.

Expecting AI to Read Your Mind

AI cannot infer what you're imagining. If you want a specific structure, mention it. If you want certain points included, list them. If you want a particular length, say so. If you're picturing a casual tone, say that.

People often leave out the most important detail because it seems obvious to them. It's obvious to you because you've been thinking about it. AI hasn't. Spell it out.

Copying AI Output Without Reviewing It

AI makes mistakes. It can be confident and wrong. It can sound authoritative while missing a key fact. It can invent things that seem plausible but aren't true.

Always read AI's output carefully before using it. Check facts. Make sure it sounds like you (if you're sending it in your name). Make sure it doesn't include anything you wouldn't say or don't agree with.

Think of AI output as a very good first draft from a very fast assistant. You still need to review it. The assistant is fast, not infallible.

Try It: Five Prompts to Practice Right Now

Reading about prompting is like reading about swimming. At some point, you need to get in the water. Here are five exercises you can do right now, with any AI chat tool, to practice what you just learned.

Exercise 1: The Specific Email

Write a prompt that asks AI to draft an email thanking someone for their time after a job interview. Include these specifics: the company name, the role you interviewed for, one thing you enjoyed discussing, and a word limit of 100 words.

What to expect: A polished, interview-appropriate email that hits all your points without rambling. If you leave out the specifics, you'll get a generic thank-you note that could apply to any interview. With the specifics, you'll get something personal and memorable.

Exercise 2: The Explanation

Ask AI to explain a concept you've always found confusing -- compound interest, how WiFi works, why the sky is blue, whatever you like. Use the CRAFT framework: give context about your background, specify that you want an analogy, and set the format.

What to expect: An explanation that actually clicks because it's tailored to how you think, not a textbook definition. The analogy makes the difference. If AI gives you one that doesn't work, ask for another analogy.

Exercise 3: The Comparison

Pick something you're genuinely deciding between -- two phones, two vacation destinations, two lunch spots, anything. Ask AI to compare them, specifying your priorities and situation.

What to expect: A comparison that highlights the tradeoffs that matter to you, not just a generic feature list. If you tell AI you care about camera quality and battery life more than gaming performance, the comparison should weight those factors.

Exercise 4: The Iteration Test

Write a prompt asking AI to suggest a dinner recipe. Get the first response, then refine it at least twice. Try: "Make it faster to cook," "I don't have [ingredient], swap it out," or "Make it kid-friendly."

What to expect: You'll see how each round of feedback improves the result. The first recipe might be fine, but the third version will be adapted to your actual kitchen and life. This exercise proves that iteration is worth the extra effort.

Exercise 5: The Question Reversal

Instead of writing a detailed prompt, ask AI to interview you first. Try: "I want to write a bio for my professional website. Ask me 5 questions before you write it." Answer the questions, then let AI write the bio.

What to expect: A bio that includes details you wouldn't have thought to include in your prompt, because AI asked about them. This exercise demonstrates why letting AI ask questions is so powerful -- it surfaces information you didn't know was relevant.

Key Takeaways from Module 3

The quality of what you get from AI depends on the quality of what you give it. Bad prompts produce bad results. Good prompts produce good results. This is not complicated, but it is important.
The five rules: be specific, give context, show examples, iterate, and break complex tasks into smaller steps. These apply to every prompt you'll ever write.
The CRAFT framework gives you a reusable structure: Context, Request, Audience, Format, Tone. You won't always need every letter, but when a prompt isn't working, CRAFT helps you figure out what's missing.
The most useful everyday prompts -- summarize, explain, draft, compare, plan, pros and cons -- follow predictable patterns. Learn the templates and adapt them.
Advanced moves like asking AI to question you, chaining prompts, and using AI to critique its own work are simple techniques that most people never try. They take very little extra effort and produce much better results.
The most common mistakes are being vague, accepting the first response, skipping context, expecting mind-reading, and not reviewing output. Now that you know them, you'll spot them.
Prompting is a skill, not a talent. Like any skill, it gets better with practice. The five exercises above are a starting point.

What's Next

You now know how to talk to AI. In Module 4 -- "AI for Your Daily Life" -- we'll put that skill to work. You'll see exactly how to use AI for the things you do every day: writing emails, planning meals, managing your schedule, learning new things, making decisions, and handling the dozens of small tasks that eat up your time.

Everything you learned here becomes more powerful when you apply it to real situations. Module 4 is full of them.