Lesson 4: Advanced Techniques That Sound Fancy but Aren't
Estimated Time: 13 minutes
Learning Objectives:
- Understand and use chain-of-thought prompting (getting AI to "think step by step")
- Use few-shot prompting (giving AI examples to copy)
- Learn 3 more power techniques: persona stacking, iterative refinement, and negative prompting
You're ready for the next level
These techniques have fancy names. Researchers write papers about them. People on Twitter make them sound complicated.
They're not.
They're simple in practice. And they're the difference between "good" and "wow, that's exactly what I needed."
Let's go.
Technique 1: Chain of Thought - "Think step by step"
What it is: Asking AI to show its reasoning before giving an answer.
When to use: Complex questions, analysis, math, decision-making, anything where you need to see the logic.
How: Just add "Think through this step by step" or "Walk me through your reasoning" to your prompt.
Example without chain of thought:
"Should I lease or buy a car?"
AI response:
"It depends on your financial situation and how long you plan to keep the car. Leasing offers lower monthly payments but you don't own the vehicle. Buying costs more upfront but builds equity."
Generic. Not helpful.
Example WITH chain of thought:
"I'm deciding between leasing and buying a car. My budget is $400/month, I drive 12,000 miles per year, and I like having a new car every 3 years. Think through this step by step, considering total cost over 5 years, and give me a recommendation."
AI response:
"Let me break this down step by step:
- Your priorities: You value having a newer car and prefer to switch every 3 years. That's a point in favor of leasing.
- Mileage: 12,000 miles/year is within typical lease limits (usually 12,000-15,000), so no penalty there.
- Cost comparison over 5 years:
- Leasing: $400/month × 60 months = $24,000. At the end, you own nothing but you've driven two different new cars.
- Buying: $400/month × 60 months = $24,000 paid toward ownership. After 5 years, you own a car worth roughly $12,000-15,000 (depending on the model).
- Your stated preference: You like having a new car every 3 years. If you buy, you'd need to sell/trade after 3 years, which adds hassle.
Recommendation: Lease. Your priorities (new car every 3 years, predictable costs, low hassle) align perfectly with leasing. Yes, you're not building equity, but since you don't plan to keep cars long-term anyway, that equity wouldn't matter much to you."
See the difference? The AI showed you its work. You can follow the logic. You can spot if it missed something. You can trust the recommendation because you saw how it got there.
Why it works: Forces AI to reason through the problem instead of jumping to a surface-level answer. The output is more accurate and more useful.
Technique 2: Few-Shot Prompting - "Here are examples, now do the same"
What it is: Showing AI 2-3 examples of what you want before asking it to create one.
When to use: When you need a specific style, format, or tone that's hard to describe in words.
How: Give 2-3 examples, then say "Now create one like these for [your topic]."
Example:
You want product descriptions for your online store, but you have a very specific style in mind. Instead of trying to describe it, you show it.
Prompt:
"Here are 3 product descriptions I like:
- The Morning Ritual Mug - This isn't just a mug. It's the vessel for your first quiet moment before the world wakes up. Double-walled ceramic keeps your coffee hot while you ease into the day. Holds 12oz of whatever fuels you. Microwave and dishwasher safe, because life's too short for high-maintenance mugs.
- The Focused Work Candle - Smells like productivity (if productivity smelled like cedar and black pepper). Light it when you need to lock in. The scent is grounding without being distracting. Burns for 40+ hours. Made with soy wax and cotton wicks. Comes in a reusable glass jar that looks good on your desk.
- The End-of-Day Journal - Five minutes before bed, that's all it takes. Guided prompts help you process the day without overthinking it. Hardcover, lays flat, fountain pen friendly. 90 days of entries. Think of it as a mental reset button you press every night.
Now write one in the same style for my product: a minimalist desk organizer made of walnut wood that holds pens, a phone, and small items."
AI output:
"The Desktop Command Center - Everything you reach for, exactly where you need it. This walnut organizer holds your pens, phone, and all the small stuff that usually clutters your desk. It's solid wood, not veneer. The grain is real. The weight feels substantial. Each compartment is sized for actual human use, not aesthetic perfection. Fits any desk, any style. Clears your space, clears your mind."
Perfect. You didn't have to explain "write in a conversational tone with short sentences, focus on benefits, make it feel premium but not pretentious, use second person." You just showed what you wanted, and AI matched it.
Why it works: Showing is 10x more powerful than telling. AI is excellent at pattern-matching. Give it a pattern to match.
Technique 3: Persona Stacking - "Respond as multiple experts"
What it is: Asking AI to respond from multiple perspectives.
When to use: Decisions, strategy, getting well-rounded advice, catching blind spots.
How: "Give me three perspectives on this: as a [expert 1], as a [expert 2], and as a [expert 3]."
Example:
"I'm thinking about starting a side business selling handmade candles. Give me perspectives from: (1) a small business accountant, (2) an experienced Etsy seller, and (3) a brutally honest friend."
AI response:
"As a small business accountant:
Your biggest challenge will be tracking costs correctly. Material costs (wax, wicks, jars, fragrance oils) add up fast, and most new makers underestimate them. You'll need to price at 3-4x material cost minimum to cover time, overhead, and taxes. Set up a separate business checking account from day one. Keep every receipt. If you make more than $600, you'll need to file a Schedule C. Don't skip quarterly estimated taxes if you're making real money.
As an experienced Etsy seller:
The market is saturated. You're competing with thousands of candle sellers. What makes yours different? If the answer is 'they're handmade and high quality,' that's not enough. Every seller says that. You need a specific angle: astrology-themed, literary quotes, a unique scent profile, eco-refillable, something. Also, budget for photography. Bad photos kill sales faster than anything. And expect Etsy fees to eat 6-8% of every sale.
As a brutally honest friend:
Do you actually like making candles, or do you like the idea of having a side business? Because making 50 candles in a weekend to fill orders is very different from making 3 candles for yourself on a Sunday afternoon. Also, how much do you need this to make? If it's 'extra spending money,' cool. If it's 'I need $1,000/month,' you're looking at 100+ candle sales per month after fees and costs. That's a real commitment. Are you ready for that?"
Now you have three angles. You can see the financial reality, the market reality, and the personal reality. Much better than a generic "sure, go for it!" response.
Why it works: One perspective can miss things. Multiple perspectives catch blind spots. You get a more complete picture.
Technique 4: Iterative Refinement - "Almost, but..."
What it is: Treating AI like a conversation, not a one-shot machine.
When to use: Always. But especially for important outputs.
How: Get the first draft, then refine it in follow-up messages.
Most people stop at the first response. That's a mistake. The magic happens in rounds 2 and 3.
Example flow:
You: "Write a LinkedIn post about why side projects matter for career growth. Keep it under 200 words."
AI: [Gives you a decent but generic post]
You: "Make it shorter. Cut it to 100 words."
AI: [Gives you a tighter version]
You: "The tone is too formal. Make it warmer and more conversational."
AI: [Adjusts tone]
You: "Good, but add a specific example in the second paragraph."
AI: [Adds example]
You: "Perfect. Now rewrite the opening sentence to be more attention-grabbing."
AI: [Final version]
You just went from "decent" to "this is really good" in 5 quick back-and-forth messages. Each refinement took 10 seconds. Total time: 2 minutes.
Compare that to trying to write the perfect prompt upfront, or writing the post yourself from scratch.
Why it works: It's faster to critique than to create. You know what you want when you see it. Use AI to generate the first draft, then shape it through conversation.
Technique 5: Negative Prompting - "Don't do this"
What it is: Telling AI what to AVOID.
When to use: When AI keeps doing something annoying in its responses.
How: Add "Don't [thing]" to your prompt. Be specific about what you don't want.
Examples:
"Write a cover letter for a marketing role. Don't use buzzwords like 'synergy' or 'leverage.' Don't start with 'I am writing to apply for.' Sound like a real human, not a corporate template."
"Explain blockchain to me. Don't use technical jargon. Don't assume I know what hashing or nodes are."
"Give me blog post ideas. Don't suggest anything that's been done to death, like '10 productivity hacks' or 'morning routines of successful people.'"
"Rewrite this email. Don't make it longer. Don't add corporate filler phrases like 'I hope this email finds you well.'"
Why it works: Sometimes it's easier to tell AI what NOT to do than to describe exactly what you want. Constraints are just as powerful as instructions.
When to use which technique
Quick reference:
- Complex analysis or decisions → Chain of thought
- Matching a specific style → Few-shot prompting
- Getting balanced advice → Persona stacking
- Polishing output → Iterative refinement
- Fixing annoying AI habits → Negative prompting
You don't need to use all 5 every time. Pick the one that fits the situation.
But once you start using these, you'll notice your results get way better.
What you just learned
- "Think step by step" (chain of thought) dramatically improves AI reasoning on complex tasks.
- Showing examples (few-shot) is more powerful than describing what you want.
- Multiple perspectives (persona stacking) catch blind spots and give you well-rounded advice.
- The best AI results come from conversation. Refine through multiple rounds.
- Telling AI what NOT to do (negative prompting) is just as useful as telling it what to do.
These techniques aren't complicated. You can start using them today. You'll immediately see the difference.
Try It Challenge
Try chain of thought right now.
Think of a question you've been mulling over. A decision you're stuck on. Something more complex than "what should I eat for dinner."
Ask AI the question and add: "Think through this step by step before giving me your answer."
Compare it to asking the same question without that instruction.
Notice how much more thorough and useful the response is. That's the power of one simple addition.
Quiz
Q1: What is "few-shot prompting"?
A) Asking AI very short questions
B) Giving AI 2-3 examples of what you want before asking it to create something similar
C) Using AI only a few times per day
D) A technique that only works in paid AI tools
Answer: B
Few-shot prompting means giving AI a few examples to learn from before asking it to produce something. It's like showing someone "write it like THIS" instead of trying to describe the style in words. Works in any AI tool.
Q2: Why should you refine your AI response instead of accepting the first result?
A) AI always gets it wrong the first time
B) It costs less to refine
C) The first response is a starting point - refining through conversation gets you the best result
D) You need to test if AI is working properly
Answer: C
AI's first response is usually decent but rarely perfect. Just like working with a human collaborator, the best output comes from feedback: "Make this shorter," "Change the tone," "Add an example." Rounds 2-3 are where the magic happens. Most people stop too early.
Next lesson: Practice lab. You'll take 5 terrible prompts and rewrite them into great ones. This is where everything clicks.