Skip to main content
There are three fundamental techniques you can use when prompting LLMs. Understanding when and how to use each will help you craft more effective prompts.

​
Zero-shot Prompting

Zero-shot prompting means asking the model to perform a task without providing any examples. Since these models have been trained on massive datasets, their internal knowledge makes them capable of performing a large number of tasks without demonstrations. Think of it like: Asking a guitar player to play the piano, even though they’ve never played piano before. They’d apply their previous knowledge about music and instruments to figure it out. Example: Prompt: 
Classify the text into the categories of satisfied, neutral or unsatisfied.

Text: I was happy with the customer support today.
Output: 
Satisfied
When to use: Perfect for general tasks like classification, translation, and answering questions with established knowledge. Most prompts we use are, by default, zero-shot prompts.

​
Few-shot Prompting

Few-shot prompting means providing demonstrations of how to perform a task. Instead of relying only on the AI model’s general training, you give specific examples that guide the model to perform your exact task with higher quality. Think of it like: Showing a guitarist a few piano pieces before asking them to play piano for the first time. Example: Prompt: 
I was happy with the customer support today - satisfied
The product is horrible! - very unsatisfied
This is one of the best products I have ever used - very satisfied
This is such a great product! -
Output: 
Very Satisfied
When to use:
  • Complex or nuanced tasks where format matters
  • When you need consistent output structure
  • 3-4 examples typically work best
Important limitation: For complex reasoning tasks, few-shot prompting hits its limits. In those cases, combine it with chain-of-thought principles for better results.

​
Chain-of-Thought Prompting

Chain-of-thought prompting improves reasoning quality by forcing the model to externalize its thinking process, essentially making it “show its work.” This is especially useful for complex or edge-case problems where the model might otherwise hallucinate or produce logically inconsistent results. Three proven techniques:

​
1. Direct Step-by-Step Instructions

Add “Think step by step” to your prompt. This simple phrase triggers the model to break down complex problems into logical components rather than jumping to conclusions.

​
2. Provide a Logical Framework

Instead of hoping the model figures out the right approach, give it the exact framework to follow. This reduces reasoning errors by constraining the solution path. Example: Vague prompt: 
Analyze the impact of climate change on polar bear populations.
Structured prompt: 
Analyze the impact of climate change on polar bear populations using this framework:
1. Current polar bear population status
2. Climate change factors affecting Arctic habitat
3. Direct impacts (habitat loss, hunting changes)
4. Indirect impacts (food chain disruption)
5. Future population projections

​
3. XML Tags for Process Separation

Use <thinking></thinking> and <answer></answer> tags to separate reasoning from final output. This prevents the model from mixing its working process with the polished result, especially useful for complex multi-step problems. Learn more about XML tags here.

​
Quick Comparison

TechniqueBest ForComplexity
Zero-shotGeneral tasks, classification, translationSimple
Few-shotTasks requiring specific format or nuanced outputMedium
Chain-of-thoughtComplex reasoning, multi-step problemsAdvanced

​
Combining Techniques

You can combine these techniques for even better results:
  • Use few-shot examples with chain-of-thought instructions for complex reasoning tasks
  • Apply chain-of-thought principles to zero-shot prompts when dealing with edge cases
  • Structure few-shot examples using chain-of-thought frameworks for consistent reasoning