How to write an AI prompt that gets you what you want.

What is a prompt?.

A prompt is the instruction you give an AI: the question or request typed into ChatGPT, Claude, Gemini or Mistral, possibly with documents and examples. The model builds its answer from that text, and that text alone.

Hence a simple rule: the AI only knows what you tell it. It does not know your company, your customer or the tone of your emails. A vague request gets a generic answer; a precise one gets a usable answer.

Anthropic's documentation puts it with a helpful image: think of the AI as a brilliant but brand-new employee who does not yet know your norms. And it suggests a test: show your prompt to a colleague with no context on the task. If they would be confused, the AI will be too.

The 5-part method.

A good prompt combines five elements. Not all are needed for a simple question, but for real work, each one counts:

Three tips from Anthropic's documentation round out the method:

• Explain the why. Instead of a bare rule, give the reason: the AI generalizes from the explanation.
• Give examples. It is one of the most reliable ways to get the right tone, form and structure. Three to five varied examples work best.
• Separate the blocks. When a prompt mixes instructions, documents and examples, marking them clearly (with headings or tags like <document>) avoids confusion.

6 business prompt examples.

Copy-ready prompts, to adapt with your own details in brackets.

1. Follow up on an unpaid invoice

You are the collections manager at a small company. Write a reminder email
for invoice [number] of [amount], [number] days overdue.
The customer is [context: loyal, new, already reminded...].
Polite but firm, 6 lines max.
End with a concrete proposal: a payment date or a call.

More on this topic: automated invoice reminders.

2. Summarize a meeting

Here are the raw notes from a meeting: [notes].
Produce minutes in three parts:
1. Decisions made, one per line.
2. Actions, with owner and due date.
3. Open questions.
Do not invent anything that is not in the notes.

3. Review a contract

You are an in-house lawyer. Here is a supplier contract: [text].
List the clauses that deserve attention: term and renewal, termination,
penalties, liability, data ownership.
For each: quote the passage, explain the risk in one sentence.
If a usual clause is missing, say so.

4. Write a product description

Here are a product's features: [list].
Write a product page for our website, aimed at [audience].
Structure: a title, a 2-sentence hook, 4 benefits as a list,
technical specs as a table.
Here is an existing page whose tone to match: [example].

5. Reply to a customer review

Here is a customer review posted on our Google listing: [review].
Write a public reply: thank them, acknowledge the point raised without
over-justifying, offer a solution or a direct contact.
5 lines max, signed [first name], [role].

6. Prepare a sales meeting

Tomorrow I meet [name], [role] at [company], in [industry].
Here is what I know: [information, website, past exchanges].
Prepare: 3 likely challenges for this kind of company, 5 open questions
to ask, and 2 possible objections with an answer for each.

The most common mistakes.

• Staying vague. "Write an email to a customer" leaves the AI guessing everything. Give the situation, the recipient and the goal.
• Forgetting the format. Without guidance, answers are often too long. Specify length and structure.
• Cramming everything into one request. For a complex task, split it into numbered steps, or chain several prompts.
• Never showing an example. One example sentence is often worth more than a paragraph describing the tone you want.
• Copying without reading. An AI can confidently make up a figure or a reference. Every factual claim gets checked before sending.
• Pasting sensitive data without checking the terms. Depending on the tool and plan, what you type may be retained. Check your subscription's terms before adding customer data.

From prompt to tool: when to take the next step.

A prompt is perfect for a one-off task. But when the same instruction comes back every week, or each colleague has their own version, there are two next steps:

• The skill. You save the procedure once, and the AI applies it on its own when a request matches. See Claude Skills.
• The automation. The prompt runs by itself, triggered by an email, a form or a date, with nobody having to start it. See How to Automate a Business Process and our custom AI agents.

To understand what happens behind a prompt, see What Is an LLM. And to pick the tool, Which AI to Choose in 2026.

AI Prompts: what people ask us.