CRAFT is a five-part framework for getting better results from AI on high-value tasks. Context. Role. Ask. Fit. Tune.

You already do most of this when you brief a teammate. You tell them the situation. You name the perspective that matters. You make the request specific. You say who the output is for. And you stay in the loop while they work. The only thing new here is treating the AI the same way.

The rest of this section walks through each letter using one realistic task: writing the opening of a proposal to a new prospect. The same prompt gets better each time we add a letter. The point is not any one output. The point is feeling how much more useful AI becomes when you give it what it needs to help you.

An example

The scenario

You work at a custom fabrication company. Twenty years of experience in precision components for aerospace. A referral just surfaced a promising opportunity: a medical device manufacturer is standing up a new product line and is looking for a fabrication supplier. You have the precision experience. You do not have direct medical device history. You have three days to send a proposal.

You sit down to draft the opening. You type the obvious prompt:

What you get back reads like a company brochure. Three paragraphs about manufacturing excellence, commitment to quality, and decades of experience serving demanding industries. It is not wrong. It is aimed at nobody. You could send it to any prospect in any industry and it would land the same way. Which is the problem.

This is the first instinct to push past. The generic response is not the AI's limit. It is what happens when the AI has nothing to work with except the shape of the request.

This is where CRAFT comes in.

Context is the situation, background, and constraints. It is the single biggest lever for getting useful output. Without it, the model guesses. And guesses wrong.

Context includes what you are working on and why, what you have already tried or decided, and constraints the AI would never know. Timeline. Audience. Company specifics. The opportunity in front of you.

Add context to the prompt:

The output shifts. It now talks about precision. It acknowledges a move into medical devices. But it is still overcooked. It spends too long on the company and not enough on the reader.

Role tells the AI what expertise or perspective to bring. It shapes the lens for everything that follows.

Add role:

Now the voice is different. It talks to someone, not about something. It leads with what a VP of Operations actually cares about: capability, reliability, and fit for their schedule.

Role matters most when the perspective matters. For a quick task, you can skip it. For anything where the reader's point of view shapes the message, name it.

Make one clear, specific request.

“Write a proposal intro” is vague. What does an intro mean? One paragraph? A full first page? Does it include pricing framing? A statement of interest?

Sharpen the ask:

A focused ask gets a focused answer. You also save yourself a round of deletion, because the AI is not trying to write the whole proposal in one shot.

Fit shapes the output to match your actual need. Tell the AI who the output is for, what format you want, and how much detail matters.

Without Fit, you get the AI's default. Usually long, faintly formal, and aimed at nobody. Fit is where you turn the output into something that belongs in your workflow.

Add fit:

The output tightens. The tone becomes warmer. The acknowledgement of the gap is short and confident instead of apologetic.

You could stop here. Many prompts will. But the last letter is where the real lift comes from.

Tune is iteration. It works in two directions.

Before the AI writes anything, you can front-load the thinking:

The AI asks the first question. Something close to:

You answer. The referrer is a consultant the VP worked with at her last company, and she asked specifically for “a shop that takes precision seriously."

The AI uses that answer and asks the next thing:

Now you are thinking. Not just answering. The second question only made sense once the AI had the first answer. And notice what happened. You came in to get a proposal written. You are now in a conversation about what actually matters to the reader of the proposal. The prompt surfaced a gap you would not have thought to fill.

That is the hinge of this whole framework. You would not have thought to put any of that into a brief to a junior colleague either. But each answer is exactly what you need to position yourself well, and the AI pulls them out of you one question at a time.

The AI started being something closer to a thought partner.

Tune also works after the first response. "Too formal." "You missed the part about their launch timeline." "Try that again, shorter, and lead with the referrer." Most people accept the first output. The best work comes from two or three rounds of feedback.

The full CRAFTed prompt