Communicating With AI · From Prompts to Reliable Results
Making AI Think
Learn how to make AI explain, compare, and reason
Getting AI to Think, Not Just Answer
Most people use AI like a search engine: they ask a question, get an answer, and move on. But AI works much better when you treat it as a collaborator rather than a lookup tool.
In this lesson, you'll learn prompting techniques that help AI think through problems, not just generate quick responses.
Imagine you're figuring out whether to take on a new client project — the timing is tight, the budget is uncertain, and you'd have to deprioritize something else. You ask AI for a recommendation and get: "Yes, take the project — the upside outweighs the risk." Sounds reasonable. But when a colleague asks how you decided, you have nothing to show them.
Choose one
What's the most likely reason that a recommendation isn't useful enough to act on?
What changes the output here is asking the AI to show how it got to that conclusion — step by step, factor by factor. That's the difference between an answer you have to trust blindly and one you can actually evaluate.
Chain-of-Thought Prompting
Chain-of-Thought (CoT) prompting adds a reasoning instruction to your prompt — something like "think through this step by step" or "reason through each factor before concluding." Instead of jumping to an answer, the AI works through the problem visibly before landing on a conclusion.
Let's see the difference in a real decision. You're considering whether to take on the client project. First, without any reasoning instruction.
practice preview
Interactive practice
Fill in the blank
Ask AI to evaluate a decision without any reasoning instruction.
You get a recommendation — but there's no reasoning.
Now add a Chain-of-Thought instruction to the same prompt and see what changes.
practice preview
Interactive practice
Fill in the blank
Add a reasoning instruction and compare what comes back.
The second response gives you something to work with: factors, tradeoffs, and the conclusion. That's what Chain-of-Thought produces — an answer you can check.
practice preview
Interactive practice
True / False
Would adding a Chain-of-Thought instruction improve the response here?
Add a Chain-of-Thought instruction when the task involves competing factors, tradeoffs, or decisions where the path to the answer matters as much as the answer itself.
Zero-Shot Chain-of-Thought
You don't always need a detailed reasoning instruction. Sometimes this short phrase is enough: "Let's think step by step."
This is zero-shot Chain-of-Thought — no examples or complex framing. Add "Let's think step by step" to the end of your prompt to trigger the same structured reasoning.
Particularly effective for calculations, comparisons, and multi-step logic.
When Chain-of-Thought Still Falls Short
Chain-of-Thought prompting improves how the AI reasons. But sometimes the problem is different: the AI reasons confidently from the wrong starting point. You can see that the logic is sound, but the assumptions just don't match your situation.
This is where a different technique comes in. Before asking your question, you ask the AI to surface what it actually knows about the topic first. That knowledge becomes the basis for the answer — and the output quality difference is hard to miss once you've seen it.
Generated Knowledge Prompting
Generated Knowledge prompting is a two-step approach:
- Step 1: Ask the AI to generate relevant knowledge about your topic.
- Step 2: Use that knowledge to answer your actual question.
Why this matters: AI tools draw on broad patterns from training data. On niche, personal, or nuanced topics, those patterns can be generic or slightly off.
Generating the knowledge first forces the AI to surface what's actually relevant to your context before it commits to an answer.
practice preview
Interactive practice
Fill in the blank
Ask the AI to surface relevant knowledge before answering an actual question.
Now take that knowledge and use it as the foundation for an actual question.
practice preview
Interactive practice
Fill in the blank
Feed the generated knowledge back in and ask your real question.
Say you're trying to decide whether to take a promotion that comes with more money but also more stress and less time. A direct prompt returns: "Weigh the financial benefits against your personal priorities and work-life balance."
But ask the AI to first generate what factors typically determine whether a promotion is the right move — ambition stage, current energy levels, family situation, growth ceiling in the current role — and the follow-up answer starts being a framework you can actually apply to your specific situation.
Select all that apply
Which of these tasks would most benefit from Chain-of-Thought or Generated Knowledge prompting?
You've now worked through techniques that change how AI processes your request. Chain-of-Thought, zero-shot Chain-of-Thought, and Generated Knowledge each target a different point in the reasoning chain — and together they give you real control over the quality of thinking behind every answer.
- Chain-of-Thought prompting makes reasoning visible: Add an instruction to think through factors to any complex prompt, and the AI shows its work before concluding.
- Zero-shot CoT works with three words: "Let's think step by step" triggers structured reasoning without any examples.
- Generated Knowledge prompting improves what the AI reasons from: Ask it to generate relevant context first, then use that to answer your actual question.
What's Next?
You can now shape outputs, control format, and guide reasoning. Up next is the final lesson of this course! You'll learn how to put all of it together — building multi-step workflows that take a problem from start to finish without losing the thread.
