The 3-part prompt: role, task, format

Type “Write about marketing” into an AI chat assistant and press send. You will get a long, tidy essay. Nothing in it is wrong. Yet you cannot post it, send it or present it. You would have to start again.

Many people decide the AI is bad at their job. That verdict is usually mistaken. Look at what you sent: four words. The AI had to guess your reader, your length and your purpose. It guessed, and none of the guesses were yours.

Three questions the AI cannot ask

Imagine a new colleague on her first morning. You say, “Write about marketing.” She would ask who she should be, what exactly you need, and what shape the answer should take. The AI has the same questions. It just does not say them out loud.

Each question has a one-sentence answer, and each answer has a name.

• Role: whose expertise and tone the AI should borrow. “You are a B2B marketer.”
• Task: the exact job, with a number. “Write three LinkedIn hooks for our invoicing app.”
• Format: the shape of the answer. “A numbered list, under twenty words each.”

None of these sentences makes the prompt long. Each one simply closes a gap. Use them in that order and the prompt almost writes itself.

Why the number in the task matters

A task without a number has no finish line. Ask for “some hooks” and you might get two or twenty. Ask for three and the AI knows when to stop. The same goes for the format. If your answer is headed for a slide, ask for lines short enough to fit on one. You save editing time later.

The role does quiet work too. It changes the words the AI picks, the examples it reaches for and how formal it sounds. A role is not a costume for show. It is a way to steer the voice.

Rebuilding the vague prompt

Start again with “Write about marketing” and add one part at a time.

• Add the role: “You are a B2B marketer.”
• Add the task: “Write three LinkedIn hooks for our invoicing app.”
• Add the format: “A numbered list, under twenty words each.”

Read it as a whole. It is one clear request instead of a vague wish. Send it, and you should get three short hooks in a list, each ready to edit. The exact words will differ every time you run it. That is normal.

Put the two results side by side. The first prompt gave you an essay to cut down. The second gave you something usable within a minute. The AI did not get cleverer in between. It was the same tool, given the same topic, a minute apart. You simply gave it fewer things to guess.

The same method in other jobs

Nothing here is special to marketing. Take a recruiter who types “Write a job ad” and gets an advert that could belong to any company. She tries again: “You are a recruiter at a small design studio. Write a job ad for a junior designer. Format: five bullets and a one-line summary.” Now the advert has a studio, a role and a shape.

Take a teacher who types “Help me with fractions” and gets a long explanation aimed at nobody. He tries: “You are a primary school maths teacher. Write four practice questions on adding fractions. Format: a numbered list, with the answers at the bottom.” This time the questions suit a real class and he can print them.

Different jobs, different words, the same three parts. When a vague answer turns up in your own work, read your prompt back and point to each part. If you cannot find one, you have found the gap.

A small habit to keep

Before you send a prompt, spend ten seconds on a quick check. Is there a role? Is there a task with a number? Is there a format? It costs almost nothing, and it saves the slow loop of sending, reading, sighing and starting over.

The sentence to remember is simple. A vague answer usually signals a vague prompt, so give the AI a role, a task and a format. From there you can add context, examples and limits to sharpen things further. But these three parts are the base that everything else sits on.