Anatomy of a Good Prompt

Two people ask ChatGPT the same question on the same morning. The first gets a lukewarm paragraph they’ll toss; the second gets text they send as-is. Neither the model nor luck separates them: only the wording changes. You build a prompt, and you can learn the rules of that construction quickly. This lesson lays out the structure OpenAI’s documentation currently recommends, which the following lessons will detail one by one.

The four elements of a prompt

ChatGPT documentation describes a prompt as a question, instruction, or objective written in your own words: there is no technical syntax or mandatory formula. For a simple request, a single sentence often suffices. For more significant tasks, it advises including the elements that matter from these four:

• The goal: what ChatGPT must do, driven by an action verb (write, analyze, compare, summarize, translate). This is the only truly indispensable element.
• The context: the information and sources that help, meaning everything ChatGPT cannot guess about you, your reader, or your file.
• The output: the expected format, length, and level of detail.
• The constraints: what must remain unchanged, what to avoid, and what ChatGPT must verify with you before acting.

Use only the elements you need. Knowing them mainly helps you spot the missing piece when an answer disappoints: a generic response often betrays missing context, while a wall of text suggests an undefined output.

Describe the result, not every step

The documentation emphasizes one point: start with the result you want, not a list of steps to get there. Specify the reader or format when those details change what ChatGPT must produce. Describe a method only when the method itself matters; otherwise, give ChatGPT the latitude to search, compare, and adjust its approach.

Turn these meeting notes into a short status update for the project team.
Put decisions and next steps first.

This prompt says what to produce and for whom. It doesn’t detail how to read the notes: that isn’t necessary.

The same need, written two ways

The difference is clearer seen than explained. First, here is the spontaneous formulation, the kind everyone types on a Monday morning.

Write an email to follow up with a client.

ChatGPT will produce a generic email, likely too long and with an uncertain tone: it doesn’t know what you sell or how long you’ve been waiting. Let’s rephrase the same request using the four elements.

Write a follow-up email for a B2B client in the logistics sector
who hasn't responded to our proposal in 10 days.

Context: We develop warehouse management software. The client
showed interest during a demo last week.

Output: An email ready to send, with subject line, body, and signature,
maximum 150 words, ending with a meeting proposal.

Constraints: Professional and warm tone; a courteous follow-up that
offers concrete help rather than pressure to decide.

The result is a targeted email with the right tone and length. The prompt took thirty seconds longer and avoids several rewrites. Notice the last line: it says what to do rather than only what to avoid. This is an explicit recommendation from the OpenAI Help Center, which Lesson 2 expands on.

Precision beats vagueness

The more precise the request, the better the response. “Give me content ideas” frames nothing: ChatGPT responds with what applies to everyone. “Propose 5 blog post ideas for a B2B digital marketing agency specializing in SEO, targeting SME marketing directors, with a title, angle, and three keywords for each” finally gives you something usable.

Precision applies to the words themselves, too. The OpenAI Help Center advises replacing vague descriptions with measurable instructions: “a paragraph of 3 to 5 sentences” rather than “a fairly short description.” However, precise doesn’t mean long: a prompt that piles on details unrelated to the result drowns out the essentials.

Order and separate

For prompts containing text to process, the same article recommends placing the instruction at the beginning and clearly separating the instruction from the content, for example with triple quotes or ###. ChatGPT then knows what you’re asking before reading the material and doesn’t confuse your instructions with the text to analyze.

Summarize the text below in five bullet points, from most to least important.

Text: """
[paste text here]
"""

Your turn

Measure the gap in your own work. Choose a task you regularly assign to ChatGPT. Write two versions: your usual prompt, then a prompt that specifies the goal, context, output, and one constraint. Compare the two responses side by side and note what the second saves you from editing. Keep this prompt: it will serve as the basis for exercises in the following lessons.

Official sources: Prompting (ChatGPT documentation) · Prompt engineering best practices for ChatGPT · Best practices for prompt engineering with the OpenAI API (OpenAI Help Center)

Key takeaways

• Build a prompt with four elements: goal, context, output, and constraints; use only those that help
• Describe the expected result rather than every step, unless the method matters
• Prefer measurable instructions to vague descriptions, and say what to do rather than only what to avoid
• Place the instruction first and separate it from the text to process