Notice: This page requires JavaScript to function properly.
Please enable JavaScript in your browser settings or update your browser.
Learn Chain-Of-Thought — Making AI Reason Step By Step | Core Prompting Techniques
Prompt Engineering for Work

bookChain-Of-Thought — Making AI Reason Step By Step

Swipe to show menu

For straightforward tasks — drafting a message, summarizing a document, generating a list — the model produces a response quickly and it's usually adequate. But for tasks that require analysis, structured reasoning, or decisions with multiple considerations, a quick response is often a shallow one.

Note
Definition

Chain-of-thought prompting is the technique for changing this. By explicitly asking the model to reason through a problem step by step before delivering its answer, you get responses that are more structured, more considered, and more useful for complex professional tasks.

What Chain-Of-Thought Looks Like In Practice

You don't need special syntax. You need a phrase that signals to the model that you want the reasoning, not just the conclusion:

  • Think through this step by step before giving your answer;
  • Before responding, identify the key considerations involved;
  • Walk me through your reasoning, then give your recommendation;
  • Break this problem down before drawing any conclusions.

Without chain-of-thought: Should we launch this feature for all users or run a limited beta first?

The model will jump to a recommendation — possibly a reasonable one, but arrived at without visible reasoning.

With chain-of-thought: Should we launch this feature for all users or run a limited beta first? Before answering, reason through the key trade-offs involved — risk, speed of learning, support load, and rollout reversibility. Then give your recommendation.

The model will surface the trade-offs explicitly before landing on a recommendation — which gives you something to react to, push back on, or use as the basis for a team discussion.

Screenshot description: Two chat panels side by side. Left panel labeled "Without chain-of-thought": user asks What's the best way to structure a performance review conversation with an underperforming employee? → AI responds with a quick bulleted list of five general tips (e.g. "be specific," "focus on behavior not personality") — correct but shallow. Right panel labeled "With chain-of-thought": same question with Think through this step by step — consider the employee's likely emotional state, the manager's goal, the legal and HR considerations, and the desired outcome. Then give a structured approach. added → AI responds with a clearly reasoned multi-paragraph response that works through each dimension before presenting a structured conversation framework. The right panel response is visibly longer and more substantive. Annotation on the right panel: "Reasoning made visible — easier to evaluate and act on."

Where Chain-Of-Thought Adds The Most Value

This technique is worth using when:

  • You're asking the model to make a recommendation or decision with multiple competing factors;
  • You need the model to analyze something critically — a proposal, a plan, a piece of writing — rather than just describe it;
  • You're using AI to prepare for a conversation or meeting and want to think through the angles in advance;
  • The task involves weighing trade-offs where the conclusion depends on how the factors are balanced;
  • You want output you can present to others — showing reasoning makes the output more credible and easier to discuss.

A Useful Variation: Ask For The Reasoning Separately

Sometimes you want the final answer in a clean format, but you also want to see the reasoning that led to it. You can ask for both explicitly:

Analyze the following proposal for potential risks. First, reason through each section and identify concerns. Then give me a summary of the top three risks in bullet points.

This gives you the structured output you need for a document or presentation, plus the full reasoning you can review — or share with stakeholders who want to understand the thinking behind the conclusions.

question mark

In which situations should you use chain-of-thought prompting to get the most value from an AI model?

Select all correct answers

Everything was clear?

How can we improve it?

Thanks for your feedback!

Section 2. Chapter 3

Ask AI

expand

Ask AI

ChatGPT

Ask anything or try one of the suggested questions to begin our chat

Section 2. Chapter 3
some-alt