Chain of Thought: How to See Inside AI's Thinking Process
How and why to ask AI to think step by step
Imagine asking two friends for advice about changing jobs. The first immediately responds with: "Yes, do it." The second takes a moment and says: "Let's think about this... considering your current situation, the job market, and your long-term goals, I think you should make the change because X." Which response gives you more confidence?
This difference between receiving a simple "do it" versus understanding the reasoning behind an answer is exactly what makes conversations with AI valuable. And that's precisely what we'll explore today with Chain of Thought – one of the most powerful techniques for improving our AI interactions.
If you've been following along, we've already explored various ways to talk to AI (also known as prompting techniques). Today, we're diving into one of my favorites: how to make AI show us its thinking process*, step by step.
*Quick note: when I say AI "thinks" or "reasons," I don't mean it actually does so like humans (not yet, anyway). In reality, it's following patterns it learned during training. It's like calling your smartphone "smart" – it's not really intelligent, it's just very good at following complex instructions.
Why Understanding AI's Reasoning Matters
When using ChatGPT, Claude, or any AI assistant, we often fall into the trap of just looking for quick answers. It's like our Google search habits – sometimes we just want a simple "yes" or "no," a direct response, something concrete. And sure, sometimes that's all we need. But there are situations where we need much more.
Think about the last time you made an important decision. Did you settle for someone simply telling you what to do, or did you want to understand the why behind it? (I bet you might have even made a pros and cons list). The same principle applies when working with AI.
This is where Chain of Thought enters the picture – a technique that lets us see the AI's reasoning process step by step. It's essentially asking AI to "think out loud" while it solves a problem or answers our question.
Better yet, when we ask AI to "think out loud," we typically get more accurate and reliable answers. It's almost as if explaining its reasoning process forces the AI to be more careful with its conclusions.
There's another crucial benefit: AIs like ChatGPT sometimes hallucinate – meaning they can invent convincing-sounding information that simply isn't true. When we can see its thinking step by step, we're much better at spotting when the AI might be making things up or venturing into unreliable territory.
The learning aspect is powerful too. When you observe how AI breaks down a problem, you're not just getting an answer – you're gaining an analytical methodology you can apply yourself in similar situations. It's like having a mentor who doesn't just hand you solutions but teaches you their approach to problem-solving.
Lastly, there's an immediate practical advantage: the ability to course-correct. When you can see the entire thought process, you can pinpoint exactly where you'd like the AI to consider different factors or take an alternative approach. Instead of just getting a flat "no" as an answer, you can say "I see you considered factor X, but how would your analysis change if we prioritized Y instead?"
What is Chain of Thought?
Chain of Thought is a prompting technique that does exactly what its name suggests: it asks AI to show us its thinking chain. Rather than just accepting the final answer, we ask the AI to explain how it arrived at that conclusion.
It's like when your math teacher insisted you show all your work, not just the final answer. (And now we understand why they did that, right? 😉)
Let's see the difference with a real-world example:
Without Chain of Thought:
User: "Should we launch our marketing campaign on Instagram or TikTok?" AI: "You should focus on TikTok for this campaign." With Chain of Thought:
User: "Should we launch our marketing campaign on Instagram or TikTok? Explain your reasoning step by step."
AI: "Let's analyze this by parts:
First, let's consider the target audience of each platform:
Instagram: Primarily millennials and professionals
TikTok: Predominantly Gen Z and younger millennials
Then, the content format:
Instagram: Photos and short videos, more curated and professional
TikTok: Short videos, more spontaneous and viral trends
Engagement and usage time:
TikTok has higher engagement rates currently
Users spend more time on TikTok than on Instagram
Based on these factors, I recommend TikTok for the campaign. The short video format and high engagement offer better opportunities to reach new audiences."Do you see the difference? The second response doesn't just tell us what to do; it lets us understand and evaluate each step of the reasoning.
How is Chain of Thought Different from Other Prompting Techniques?
If you've been following this series on prompting techniques (and if not, no worries!), you might be wondering what makes Chain of Thought special compared to Zero-Shot Prompting or In-Context Learning.
Here's an analogy that helps clarify the differences:
Zero-Shot is like asking someone directly: "Where's the subway station?"
In-Context Learning is like showing examples first: "Here's how others have given directions..."
Chain of Thought is like saying: "Walk me through, step by step, how to get to the station"
Let's break this down further:
Zero-Shot vs Chain of Thought:
With Zero-Shot, we simply ask our question and hope the AI understands what we want
With Chain of Thought, we specifically ask to see the AI's thinking process
In-Context Learning vs Chain of Thought:
In-Context Learning provides examples to help the AI understand what we're looking for
Chain of Thought focuses on making the reasoning visible, with or without examples
What's fascinating is that these techniques aren't mutually exclusive. Think of them as different tools in your toolbox – you can use them individually or in combination:
You can provide examples of step-by-step reasoning (combining In-Context Learning with Chain of Thought)
Or simply ask for the thinking process without any examples (combining Chain of Thought with Zero-Shot)
The key difference is that Chain of Thought isn't about how AI learns or interprets your question – it's about how it reveals its thinking to you. It's the difference between someone handing you an answer versus walking you through how they figured it out.
How to Use Chain of Thought
We now understand what Chain of Thought is and why it's valuable. But the crucial question remains: how do we actually get AI to show us its thinking process?
You might assume simply saying "explain your reasoning" would be enough. And technically, it would work... but that's like asking someone to explain how they got home without specifying which part of the journey you're interested in. You'll get an answer, but probably not the most helpful one.
The secret lies in being specific and structured. Think of it as guiding the AI's thinking along a particular path. Here's how to do it effectively:
The Basics:
Start with these simple yet powerful phrases:
"Let's analyze this step by step..."
"Let's break down this problem..."
"Show me your thinking process as you..."
The Advanced Approach:
Rather than just requesting reasoning, actively guide that reasoning. It's like providing someone with a map instead of just telling them to reach a destination. For example:
"Analyze this situation considering:
The immediate impact
The long-term consequences
The necessary resources
For each step, explain why that factor is important to the final decision."
The best part? As you start using Chain of Thought regularly, you'll begin to recognize patterns in AI responses. You'll discover which approaches get better results, allowing you to fine-tune your prompts for even more effective interactions.
Common Mistakes to Avoid:
Asking for reasoning after receiving a response - it's like asking for a recipe after you've finished eating the meal
Being too vague - AI is smart, but it can't read your mind
Not specifying which aspects matter most - remember, you're in the driver's seat of the conversation
Wrapping Up
Chain of Thought is a simple yet powerful technique. You won't need it every time – sometimes a quick, direct answer is exactly what you're after. But when you want to understand the reasoning behind a recommendation or need deeper, more reliable answers, you now have this valuable approach in your toolkit.
This technique is just one piece of our prompting arsenal. In upcoming posts, we'll explore additional techniques that will help you get even more value from these AI tools.
Thanks for reading!
G
Hey! I'm Germán, and I write about AI in both English and Spanish. This article was first published in Spanish in my newsletter AprendiendoIA, and I've adapted it for my English-speaking friends at My AI Journey. My mission is simple: helping you understand and leverage AI, regardless of your technical background or preferred language. See you in the next one!