The 'Show Your Work' Habit

Remember Mrs. Higgins from 5th-grade math?

You’d stare at a long division problem, scribble the answer "42" at the bottom, and hand it in. And what did she do? She marked it wrong.

"I don't care if the answer is right," she'd say, tapping the paper with her red pen. "You didn't show your work."

We all hated it. But here is the irony: Mrs. Higgins was the first Prompt Engineer.

It turns out, Artificial Intelligence works remarkably like a 5th grader. If you let it jump straight to the answer, it makes mistakes. But if you force it to "show its work," it becomes a genius.

The Problem: AI is a Guesser

We tend to think of AI as a calculator—a cold, hard logic machine.

It's not. It is a Pattern Matcher.

When you ask a question like "Alice has 4 brothers. Each brother has 2 sisters. How many sisters does Alice have?", the AI might just do a quick math calculation (4 brothers x 2 sisters = 8 sisters).

Wrong.

It didn't pause to realize that Alice herself is one of the sisters, and all the brothers share the same two sisters. (The answer is 2, by the way).

Often, it guesses wrong because it skipped the logic and went straight for the prediction. It didn't "think." It just blurted out an answer.

The Solution: Chain of Thought

In the nerdy world of AI research, the fix for this is called Chain of Thought (CoT) prompting.

It sounds fancy, but it's literally just asking the AI to narrate its inner monologue.

Instead of asking:

"Q: Roger has 5 tennis balls. He buys 2 more cans of tennis balls. Each can has 3 balls. How many tennis balls does he have now?"

You ask:

"Q: Roger has 5 tennis balls. He buys 2 more cans of tennis balls. Each can has 3 balls. How many tennis balls does he have now? Let's think step by step."

Those five words—"Let's think step by step"—are magic.

They force the AI to stop guessing and start reasoning. It produces text like this:

1. Roger starts with 5 balls.
2. 2 cans x 3 balls per can = 6 new balls.
3. 5 + 6 = 11.
4. The answer is 11.

By generating those steps, the AI creates its own logic path to follow, drastically reducing errors.

The Standard Prompt vs. Chain of Thought (CoT)

Here is how adding "Let's think step by step" changes the result:

Task	Standard Prompt Result	Chain of Thought Result
Math / Logic	Guesses the answer (often wrong).	Calculates the steps (usually right).
Coding	Spits out code immediately (often buggy).	Plans the logic first, then writes code.
Strategy	Gives generic advice.	Outlines constraints, then gives specific advice.
Writing	Writes a draft immediately.	Outlines the main points first, then writes.

Example: The "Content Strategy" Fail

This applies to more than just math riddles. It's crucial for writing and strategy, too.

The "Lazy" Prompt:

"Write a blog post about why remote work is good."

The Result: a generic, vanilla article about "saving commute time" and "working in pajamas." (Yawn.)

The "Show Your Work" Prompt:

"I want to write a blog post about why remote work is good.

Step 1: Brainstorm 3 unique, contrarian angles that most people ignore. Step 2: Select the most compelling one and outline the main arguments. Step 3: Write the blog post based on that outline."

The Result: The AI might choose an angle like "Remote work is actually better for extroverts because focused time makes social time more meaningful." Then it writes a brilliant, specific piece.

By splitting the task, you gave the AI space to "think" before it started "talking."

The Counterpoint: "Don't New Models Do This Automatically?"

You might be thinking: "Wait, don't the newest models like OpenAI's o3 or Google's Gemini 3 Pro (see The Spec Sheet) already 'think' before they answer?"

Yes, they do. These "reasoning models" have Chain of Thought built right into their architecture. They spend time processing (showing their work invisibly) before they give you an answer.

So, is manual Chain of Thought dead?

Not at all. While o3 handles logic automatically, it doesn't know your strategy.

• o3 can figure out the math puzzle alone.
• o3 cannot guess your specific content strategy constraints unless you force it to outline them first.

Thinking "step-by-step" isn't just for error correction anymore; it's for control. Using a "Scratchpad" step allows you to review the AI's logic before it commits to the final work, saving you from editing a 2,000-word mistake.

Your New Habit

Whenever you have a complex task—whether it's a logic puzzle, a coding problem, or a strategic plan—don't just ask for the deliverable.

Ask for the Scratchpad.

• "Outline the logic before you write the code."
• "List your assumptions before you give the recommendation."
• "Think step by step."

Make Mrs. Higgins proud. Show your work.