CHAIN OF THOUGHT

Tell Claude to think step-by-step BEFORE answering and the answer gets meaningfully better. The technique researchers use to push AI past surface-level responses. One sentence. Big difference.

THE PATTERN

Add to any prompt:

Think step-by-step before answering.

Or the longer version:

Before you answer, think through this carefully:
1. What is the user actually asking?
2. What context matters?
3. What are the possible approaches?
4. Which is best, and why?

Then give your final answer.

WHY IT WORKS

AI models give better answers when forced to reason explicitly. This isn't a Claude trick — it's a property of how language models work. Letting the model "talk through" the problem improves accuracy on:

Math
Logic puzzles
Complex multi-step decisions
Code (especially debugging)
Anything where the wrong first instinct leads to a wrong answer

CHAIN OF THOUGHT vs. SHOW YOUR WORK

These overlap but aren't identical:

"Show your work" (Day 14) = output the reasoning
"Chain of thought" (Day 17) = think the reasoning, then output

Often you want both: think it through, then show me. Combined version:

Think step-by-step before answering. Then show me your reasoning AND the final answer.

WHEN TO USE IT

✅ Use for:

Math problems where the obvious answer might be wrong
Decisions with non-obvious tradeoffs
Code with edge cases
Strategy / business analysis
Anything you'd benefit from a second opinion on

❌ Skip for:

Quick lookups
Creative writing
Pure information requests
Anything where speed matters more than precision

REAL EXAMPLE

Without CoT:

A train leaves Chicago at 2pm going 60mph. Another leaves NYC at 3pm going 70mph. Chicago to NYC is 800 miles. When do they meet?

Possible answer: a quick (potentially wrong) number.

With CoT:

A train leaves Chicago at 2pm going 60mph. Another leaves NYC at 3pm going 70mph. Chicago to NYC is 800 miles. When do they meet?

Think step-by-step. Set up the problem, identify the variables, work through the math. Then give the answer.

You'll see Claude set up the closing-distance equation, account for the 1-hour head start, and get the right answer.

ADVANCED: SELF-CHECKING CoT

Think step-by-step. After you have an answer, check it:
1. Does the answer make sense given what was asked?
2. Are there edge cases the answer breaks on?
3. Is there a simpler answer I missed?

Then give the final answer.

This is how you get answers that don't fall apart on second look.

3 THINGS YOU CAN'T SKIP

1. Don't use CoT for everything. It's overkill for "what's the capital of France." Match the technique to the stakes.

2. Read the chain. If Claude reasons wrong in step 2, the answer is wrong. The point of CoT is verifiability.

3. Combine with magic words for deep work: "Ultrathink. Step-by-step. Be brutal. Final answer at the bottom."

WHAT'S NEXT

Day 17 of 100. Tomorrow: Day 18 — The "Critique Your Own Answer" Pattern.

SAFETY CHECK

Same as Day 1.