Practice10 min read

Chain-of-Thought: Make AI Show Its Work

The single most underused trick in prompting

Ask AI a multi-step question and it'll often leap to an answer without showing how it got there. Sometimes that answer is right. Sometimes it's confidently wrong. You can't tell which.

The fix is one sentence: ask AI to think step by step before giving its answer. This is called chain-of-thought prompting, and it dramatically improves accuracy on reasoning tasks — at the small cost of a longer response.

Three ways to invoke it

The basic version: "Think step by step before answering."

The structured version: "Walk me through your reasoning in numbered steps. Then state your final answer."

The diagnostic version: "Before answering, list the factors you're weighing and how each one points. Then give your conclusion."

The structured and diagnostic versions are better for high-stakes questions where you actually need to check the reasoning, not just have it on the page.

When NOT to use it: simple factual lookups, short edits, brainstorming. Chain-of-thought adds tokens and reading time. Use it for analysis, multi-step decisions, math, or any time you'd want to challenge AI's logic afterward.

Practice Exercise

Here's a basic prompt that's getting weak results: > "Should we offer free shipping to our customers?" The answers AI gives are generic ("it depends on margin," "consider customer experience") and you can't tell if AI actually thought about your situation. Your context: you're a 30-person DTC company selling premium skincare ($45-90 products), current AOV is $68, current shipping cost averages $7, your direct competitors all offer free shipping above $50. Rewrite the prompt using chain-of-thought to get a defensible answer.

No pressure — just give it your best shot! Write a prompt for the scenario above and our AI will give you friendly, specific feedback on how to improve.

0 / 5,000

(try writing your own prompt first!)