Prompting Fundamentals

What is prompting?

Prompting is writing clear instructions or questions that tell an AI system exactly what I want it to do or answer. It sounds simple, and the mechanics are, but the quality of the instruction determines the quality of what comes back.

Why does it matter?

• An AI system's output can only be as good as its input.
• Clear, well-structured prompts are the lever I have for getting accurate, relevant, and trustworthy results, especially in fields where the cost of a wrong or half-right answer is high.
• This is not a matter of taste. A field experiment run by researchers at Harvard, MIT, and Wharton with consultants at Boston Consulting Group found that teams who were given structured prompting guidelines completed more work, did it faster, and produced work judged to a higher standard than a control group working without AI.

You can read the original coverage of that study here. The finding that matters for how I work: the guidelines around prompt structure were part of what drove the gap between the group using AI well and the group that was not.

How I write a prompt

1. State what I want, up front.
   "Summarize this article in under 100 words." A clear, single-sentence request helps the model lock onto the task.
2. Add just enough context.
   I give the model the information it cannot guess: links, data, audience, or my goal. "Here is the paragraph below. Rewrite it for a general audience."
3. Specify the format I need.
   Bullet list, table, or plain text, I tell the model which one. "Return a two-column table: 'Pro' and 'Con.'"
4. Set the tone or role, if it matters.
   Style cues work, but I keep them short. "Write in a direct, professional voice."
5. Use examples for tricky tasks.
   I show one or two input-to-output pairs. The model will follow the pattern.
6. Bundle instructions, then send.
   AI chat platforms read the whole prompt before answering, so I group everything together rather than drip-feeding instructions, unless I actually want a back and forth.
7. Set off long inserts.
   I wrap pasted text in triple backticks or a similar marker so the model knows where it starts and ends.
8. Treat it like a conversation.
   If the result is not quite right, I tweak a sentence or add an example rather than starting over. Small edits often move the answer from adequate to good.

A working structure for a prompt

Most prompts that work well share the same underlying structure. I write out the task first, then add the context the model cannot infer on its own, then specify the format I want back.

TASK: <what you need>

CONTEXT: <info the model needs>

FORMAT: <bullet list, table, code, etc.>

TONE/ROLE: <optional style cues>

EXAMPLES (optional):
Q: ...
A: ...

TASK: Draft a polite follow-up email.

CONTEXT: I applied for the Product Manager job at ACME 10 days ago; no reply yet.

FORMAT: 5-sentence email, plain text.

TONE: Friendly and concise.

This template is a starting point, not a formula to apply mechanically. For a fuller set of practices, see Prompt Patterns: 10 Practices for Reliable Output.