Advanced Prompting Techniques

There's a phrase that costs you nothing to type and consistently makes AI output better on complex tasks. Four words:

"Think step by step."

That's it. Tack it onto any prompt where the AI needs to reason through something, and the quality jumps. Not a little. A lot.

Sounds too simple, right? Let me show you why it works.

The shortcut problem

When you ask AI a complex question in one shot, it does what any of us would do if forced to answer immediately: it takes shortcuts. It jumps to a conclusion. Sometimes that conclusion is fine. But for anything with multiple layers, that's where things go sideways.

Here's an example. Say you ask:

"Should we open a second office in Austin?"

The AI will give you a quick answer. Probably a list of pros and cons that sounds reasonable but feels shallow. It's the business equivalent of asking someone for their gut reaction.

Now try this instead:

"I'm the CEO of a 50-person marketing agency based in New York. We're considering opening a second office in Austin. Think step by step. First, analyze the key factors we should consider. Then evaluate each factor for our specific situation. Then give me your recommendation with the reasoning behind it."

Completely different output. You'll get structured analysis that actually reflects the complexity of the decision.

Why this works

Remember from Week 1 when we talked about how AI predicts the next word? When you ask it to think step by step, you're literally changing the sequence of words it generates. Instead of jumping to "Austin would be a good choice because..." it starts with "Let's consider the key factors..." Each step builds on the previous one. The reasoning accumulates.

Researchers at Google actually published a paper on this in 2022. They called it "chain of thought prompting." They found it dramatically improved performance on math, logic, and complex reasoning tasks. But you don't need to know the research. You just need to use it.

Breaking big asks into stages

The step-by-step approach has a cousin that's equally powerful: breaking your prompt into explicit stages.

Instead of one massive request, you structure it like a conversation:

Stage 1: "Here's a 10-page report on our Q3 performance. Read it and summarize the 5 most important findings."

Stage 2: "Now compare those findings against our Q3 goals. Where did we hit, where did we miss?"

Stage 3: "Based on the gaps, recommend 3 specific actions for Q4. For each, estimate impact and effort."

Each response builds on the last. The AI maintains context across the conversation (we'll talk more about how context works in the next lesson). And you can course-correct at each stage instead of getting a final output that missed the point.

Think of it like managing that brilliant intern from Week 1. You wouldn't hand them a vague brief and disappear for a week. You'd check in at each step. Same principle here.

When to use this

Not every prompt needs chain of thought. For simple stuff like "rewrite this email to be shorter" or "translate this to Spanish," just ask directly. You'd be adding friction for no reason.

Use step-by-step thinking when:

• The task involves analysis or reasoning (not just generating text)
• There are multiple factors to weigh
• You need the AI to consider tradeoffs
• The answer depends on context that's not obvious
• You tried the direct approach and the output was too shallow

A good rule of thumb: if a smart human would need to think about it for more than 30 seconds, tell the AI to think step by step.

Combine it with yesterday's lesson

Remember the persona trick from Day 8? Stack them.

"You are a senior financial analyst. I'm going to share our Q3 numbers. Think step by step. First identify any concerning trends. Then assess the severity of each. Then recommend what we should do about the top 3."

Role plus chain of thought. That's a one-two punch that handles probably 70% of the complex work you'd want AI to help with.

AI is better at thinking when you ask it to show its work. Just like humans.