How to Write an AI Prompt That Works: Context, Task, Format (2026)

You are reading Chapter 5 of 36 in the AI Bootcamp, AI Central's free course for professionals who want to use AI in their own work. The chapters build on each other, so if this is your first one, Chapter 1 is the place to start. All of them sit on the AI Bootcamp page, and three more arrive every week.

An AI prompt that works carries three things: context, one task, and a format. Vague asks get average answers because the assistant has to guess who the answer is for, what you are actually trying to get done, and what shape you want it in - and it guesses ordinary, then delivers it with confidence.

This guide gives you the three parts, a test for whether your ask has them, the same task written badly and then well, and four prompting habits you can stop typing.

We publish AI systems like this for 300,000+ senior professionals at AI Central, and more of them live in the AI Central Library.

Ask for "a summary of our Q3 results" and the assistant has three blanks to fill: who is reading this, what it is for, and how long it should be. It fills them with the most ordinary answer available, because that is the safest guess when nobody said.

Here is the test. If the answer you got would suit any company in your industry, it was written for none of them. That is not the assistant failing. That is what you asked for, answered literally.

The fix is not a longer prompt. It is naming the things you already know and did not say.

Who this is for, what moment it lands in, what the reader already knows. The company, the quarter, the room. This is the part you hold and the assistant cannot get anywhere else.

One verb, one job. If your sentence contains "and", you have probably asked for two things and will get both done at half strength.

How long, in what form, for what delivery. Six bullets. An email. A one-page brief. Unstated format is the single most common reason a technically correct answer is unusable.

A quick way to check all three at once: could someone competent, handed only these words and none of your context, do the job? If not, the missing information is what to add.

The weak version is nine words long and names none of the three parts. The strong version names the reader, the moment it lands in, and the shape it should take: six bullets, each carrying a claim and a number.

What is worth noticing is that nothing in the second version required new work. Every addition was something already in your head and simply left unsaid. The rewrite takes under a minute once you know which three blanks you are filling.

Four habits from the early prompting era no longer earn their keep on current assistants.

"You are a world-class strategist" does not make the answer better. Naming the actual reader does.

On reasoning models this is already happening. Adding the line changes nothing.

Be polite if you prefer to be. It is not a lever, and neither is rudeness.

Promising money for a better answer was always odd and does not work. Nothing is being withheld pending payment.

All four are attempts to unlock effort that is not being held back. Context, one task and a format do the work those tricks were standing in for. If you want the full catalogue of techniques that do hold up, our guide to the 26 principles of prompt engineering goes deeper.

Before: you type one line, get something that would suit anyone, and repair it three times to claw back the specifics you knew all along.

After: you name the reader, the one job and the shape before you send. The first answer arrives close enough that the repair lines from Chapter 4 are polish rather than rescue.

One thing carries into Chapter 6: a question that stayed hard even when you asked it well. Chapter 6 is about telling a genuinely difficult question from a badly worded one, and what to do when it is the former. The rest of the syllabus lives in the AI Central Library.

Three parts: context (who it is for and what moment it lands in), one task stated with a single verb, and a format (how long, what form). Most disappointing answers are missing one of the three rather than being badly written.

Because the request did not say who the answer was for or what it was for, so the assistant filled those blanks with the most ordinary version available. If the answer would suit any company in your industry, that is the reason.

No. Length is not the variable. A two-line ask that names the reader, the job and the format beats a paragraph of instructions that names none of them.

Not reliably. Expert personas, "think step by step" on reasoning models, politeness used as a technique, and offering a tip are all habits from the early prompting era that no longer earn their keep. Naming the real reader does what the persona was standing in for.

Use the colleague test: could someone competent, handed only these words and none of your context, do the job? Whatever they would have to ask you is what the assistant is guessing at.