What a prompt is and how to write one that works

The model doesn't know your company, your customer or the tone you use. Whatever you don't write in the prompt, the model guesses. That is why two people using the same tool get very different results.

A good prompt says who the reader is, what the purpose is, what shape the result should take and what to avoid. If you have an example of a good result, pasting it in is worth more than ten lines of explanation.

An example in a small business

An owner asks "write an email to a customer who is late with payments" and gets a cold, generic text. Then she rewrites it: "Write a short email to a long-standing customer, a good payer, who has an invoice that is a few weeks overdue. Friendly tone, no threats, suggest a phone call. Sign it as Laura, accounts." The second result can be sent almost without changes.

The most common mistakes

• Writing one line and expecting the model to understand the context you have in your head.
• Putting too many requests in a single prompt: it is better to do one task at a time.
• Not saving the prompts that work: in a company they become templates everyone can reuse.

Why it matters to you

Knowing how to write a prompt is the skill that makes the difference between people who find AI useful and people who find it useless. You can learn it in a few hours and apply it to any tool.

All the AI terms explained simply are in the glossary.