There is no single correct way to write a prompt. Still, most effective prompts follow a few common structures. Each one works better for certain tasks and comes with its own strengths and challenges.
In this section, we will explore the main types of prompts used in real systems:
- Zero shot
- One shot
- Few shot
- Chain of thought
- Role based
- Context rich
- Completion style
You will learn when to use each type, what it looks like, and how models respond to them.
For more practical tips and examples, check out:
2.1 Zero shot prompts
What it is
A direct instruction with no examples.
When to use it
Use this format for simple, well-known tasks.
Example
Extract three key action items from the meeting notes below. List them.
Strengths
- Simple and fast
- Easy to update
Weaknesses
- May fail on unusual edge cases
- Needs clear instructions
Model tips
- GPT-4o prefers short, clean input
- Claude 4 benefits from a bit of tone and audience framing
- Gemini responds better with section titles like “Task”
2.2 One shot prompts
What it is
This format includes one example to guide the model.
When to use it
Good when tone or format matters, and one example is enough.
Example
Input: “Hello, I need help tracking my order.”
Output: “Thanks for reaching out. I can help you track that. What’s your order number?”
Now respond to: “My package hasn’t arrived yet.”
Strengths
- Clear tone and structure
- Makes outputs more consistent
Weaknesses
- One example might not be enough
- If the example is not separated well, the model might get confused
Model tips
- GPT-4o copies format well
- Claude 4 keeps tone even if structure changes
- Gemini prefers clear “input to output” formatting
2.3 Few shot prompts
What it is
Includes several examples to teach a pattern.
When to use it
Use this when the task is tricky or needs consistent tone.
Example
Input: “Hi, I lost my password.”
Output: “Sorry to hear that. Please follow this link to reset it.”
Input: “I want to delete my account.”
Output: “We’re sorry to see you go. Here’s how to do it.”
Input: “I was double charged.”
Output: …
Strengths
- Makes the pattern clearer
- Helps avoid made-up answers
Weaknesses
- Uses more space
- If examples are wrong, output will be wrong
Model tips
- GPT-4o prefers short, clear samples
- Claude 4 works well with fewer but detailed ones
- Gemini likes examples that are spaced apart
2.4 Chain of thought prompts
What it is
The model is guided to reason step by step.
When to use it
Best for logic, math, or decisions.
Example
Let’s solve this one step at a time.
First, find the inputs.
Next, apply the logic.
Then, write the answer.
Strengths
- Improves accuracy on tough tasks
- Shows model thinking clearly
Weaknesses
- Slower to respond
- Sometimes too wordy
Model tips
- GPT-4o likes ordered steps
- Claude 4 works well with step tags
- Gemini prefers structured logic over freeform
2.5 Role based prompts
What it is
The model is told to act in a specific role.
When to use it
Use this when voice or knowledge area matters.
Example
You are a cybersecurity expert. Summarize this report for a non-technical policymaker in under 200 words.
Strengths
- Sets the right tone
- Matches audience better
Weaknesses
- Needs reminders in long prompts
- Repeating roles can sound fake
Model tips
- GPT-4o works well when role is set in system prompt
- Claude 4 responds to strong identity cues
- Gemini works best when the role is followed by a clear task
2.6 Context rich prompts
What it is
You include full background data, like a chat, doc, or transcript.
When to use it
Use when the model needs to refer to a lot of info.
Example
From the transcript below, find the key points of disagreement.
Transcript: “”” … “””
Strengths
- Helps with summarizing or complex questions
- Keeps answers grounded in real data
Weaknesses
- Models might get lost in the text
- Harder to format correctly
Model tips
- GPT-4o likes context at the top
- Claude 4 keeps memory across long inputs
- Gemini needs headers and structure to navigate large inputs
2.7 Completion style prompts
What it is
You start a format or sentence, and the model finishes it.
When to use it
Good for templates or writing that follows a structure.
Example
Incident Report
Summary:
Impact:
Resolution:
Strengths
- Easy to shape the output
- Good for form filling
Weaknesses
- Model might go off track
- Needs a clear and strong start
Model tips
- GPT-4o completes naturally
- Claude 4 keeps flow smooth
- Gemini likes clear structure cues
Summary
Prompt structure changes how models behave. It affects:
- How clear the response is
- How well the format holds up
- Whether the task works reliably
Picking the right structure is not just helpful. It decides how well your system works in the real world.
Next, we will look at advanced tricks like scaffolding, anchoring, compression, and planning.
