What Is a Prompt
A prompt is everything you send to a model before its response. A good prompt gives the model enough signal to produce what you actually need — not just what it considers the statistically likely next token.
The Four Components
Most strong prompts combine some mix of four building blocks:
| Component | Purpose | Example |
|-----------|---------|---------|
| Role / Persona | Sets tone and expertise | "You are an experienced UX editor" |
| Context | Gives the model necessary background | "We write for an audience aged 50+" |
| Instruction | Says what to do | "Rewrite this paragraph more simply" |
| Constraints & Format | Narrows the response space | "No more than 3 sentences, no jargon" |
Not every prompt needs all four — but when results disappoint, check which component is missing.
Weak vs. Strong Prompt
Task: write a product description for an online store.
Weak prompt:
Write a description for wireless headphones.
Problem: the model doesn't know the audience, tone, length, or key benefits. The result will be generic.
Strong prompt:
You are a copywriter for an electronics e-commerce store. Our audience is young professionals aged 25–35 who value sound quality and style.
Write a product card description for the Sony WH-1000XM5 wireless headphones.
Key benefits: active noise cancellation, 30-hour battery life, foldable design.
Constraints: 3–4 sentences, conversational tone, no technical jargon. End with a call to action.
The difference is clear: the second prompt gives the model a role, audience, specific product, facts, and format.
Explicitly specifying the format dramatically increases predictability:
Without format:
List the pros and cons of remote work.
The model may reply with a paragraph, a list, a table — anything.
With format:
List the pros and cons of remote work as a Markdown table with two columns: "Pros" and "Cons". Four items in each column.
Now the output is predictable and ready to use.
Iterative Improvement
Prompting is iteration. When a result disappoints, ask yourself:
- Which component is missing?
- Is the instruction specific enough?
- Would an example of the desired output help?
- Is the output format clearly specified?
Each prompt revision is a hypothesis. Test one change at a time.