Anatomy of a Good Prompt

A language model cannot read your mind: it relies only on the text you give it. A vague request produces a generic answer, while a precise request produces a useful one. The good news is that there is a simple, reusable structure that works for almost any task, and in this lesson you will learn to apply it.

In chapter 1 you saw that a model builds its answer from the text it receives. It follows that the quality of your input largely decides the quality of the output. Think of a very capable assistant who has just joined your team and knows nothing about your situation: the clearer your brief, the better their work.

Why wording changes everything

Take the request "Write something about our team meeting." The model has to guess a lot: what kind of text (a summary? an invitation? a joke?), for whom, how long, in what tone, with which information. It fills each gap with the most statistically likely choice, which is usually bland. A good prompt removes as many of these guesses as it reasonably can.

The five ingredients

1. Role: who should be speaking? ("You are a career counselor...", "Act as a patient math tutor..."). A role steers vocabulary, tone and the type of advice.
2. Context: the situation, the audience, the purpose. Who will read this? Why does it matter? What has already happened? This is the most often forgotten ingredient, and often the most valuable.
3. Task: exactly what you want, with a precise action verb (write, compare, summarize, correct, classify, list).
4. Format: the expected shape of the answer: a bulleted list, a table, 150 words, a three-paragraph email with a subject line.
5. Constraints: what to respect or avoid: tone, length, banned words, language level, and rules such as "do not invent any data."

A handy memory aid: R-C-T-F-C. You do not need to memorize the order; what matters is that you check each ingredient before pressing Enter.

Before and after

Here is a typical beginner prompt:

Write an email to reschedule a meeting.

The result will be correct but impersonal, and you will have to fill in everything. Here is the same request, structured:

Role: You are my executive assistant, courteous and concise.

Context: I need to reschedule a project follow-up meeting with an
important client. It was planned for Thursday, and an unexpected
commitment prevents me from attending. The client is flexible but busy.

Task: Write an email proposing a new time slot.

Format: A subject line, then three sentences maximum, then a closing.

Constraints: Professional but warm tone. No excessive apologies.
Leave the dates as [date] placeholders so I can fill them in.
Do not invent any reason for the change.

Each line removes a source of uncertainty. The model no longer has to guess the length, the tone, the recipient or the level of apology. Notice also the last constraint: you explicitly forbid something the model might otherwise add to make the email "sound more natural."

Writing a strong task

The task is the core of your prompt, and its wording matters.

• Use a concrete verb. "Help me with my CV" is vague. "Rewrite these three bullet points so each starts with an action verb and states a result" is precise.
• One main task at a time. If you ask for a summary, a critique, a translation and a social media post in one go, the quality of each part drops. Break it into steps (you will learn to chain prompts in chapter 3).
• Say what "good" looks like. "Make it better" means nothing. "Make it clearer for someone who has never worked in finance" gives a target.

Giving useful context

Context answers the questions the model cannot ask on its own. Useful context often includes:

• The audience: a manager, a ten-year-old, an expert, a customer who is upset.
• The goal: to inform, to persuade, to apologize, to decide.
• The situation: what happened before, what is at stake, what is already decided.
• The source material: the text to rewrite, the notes to summarize, the facts to use.

When the model must rely on facts, paste them in and tell it to use only what you provided. This is far more reliable than hoping it remembers the right details.

Specific beats clever

Many beginners search for "magic words" or elaborate formulas. In practice, plain and specific wins. Compare:

• Vague: "Give me ideas for my presentation."
• Specific: "I give a 10-minute presentation to non-technical colleagues about why we are changing our file-sharing system. Suggest five opening hooks that are friendly and under 25 words each."

The second prompt contains no trick. It simply says who, what, how long and in which style.

Common mistakes

• Too vague: "Tell me about marketing."
• Too many requests mixed together: better to split them into steps.
• Implicit context: what is obvious to you (your company, your region, your previous emails) is invisible to the model.
• Only negative instructions: "Don't be boring" helps less than "Use an upbeat tone and one concrete example."
• Contradictory constraints: "Be extremely detailed, in 50 words."
• Copying a prompt from the internet without adapting it: a prompt is a recipe for a particular situation.

Prompts are not commands to a machine

You can write to the model in ordinary sentences. There is no special syntax to learn. Politeness is not required (and it will not offend the tool), but clarity is. If you find it easier to write in the second person ("You are..."), do so; if you prefer headings such as Context: and Task:, that works too. Pick a layout that makes your own thinking clear, because a well-organized prompt is also easier for you to reread and improve.

Your turn: build your first library entry

Exercise 1. The following prompt is too vague: "Summarize this article." Rewrite it using the five ingredients for a situation of your choice (for example, a summary for your manager before a meeting).

Exercise 2 (project). Pick one of the three tasks from your prompt notebook. Write a first version of the prompt as one line, run it, and keep the answer. Then rewrite it with role, context, task, format and constraints, and run it again. Compare the two results and write down what changed.

Role: You are an assistant who prepares briefings for busy managers.
Context: My manager has five minutes before a meeting about
remote-work policies. She has not read the article below.
Task: Summarize the article so she can take part in the discussion.
Format: Five bullet points, then one sentence starting with
"Why it matters:". Maximum 120 words.
Constraints: Plain language, no jargon. If the article does not give
a figure, do not add one. Flag anything that looks uncertain.

Article: [paste the text here]