How to Write an AI Prompt That Gets a Usable Answer: The CRAFT Framework

Type “write me a cover letter” into any AI tool and you will get something so generic it could belong to anyone. That is not the AI failing you. It is the request failing.

I dedicated my career to developing and implementing training for the United States Navy. And a weak prompt fails the exact same way a weak training request fails: no audience named, no objective set, no standard given. Here is the five-part framework I use to fix that, and you do not need to be technical to run it.

Most people treat AI like a search engine. You type a few words, you get a wall of text, and you take whatever it hands back. Then you wonder why it feels generic. The reason is simple. The AI answered the question you actually typed, not the one in your head.

In training, we never send a vague request into a classroom. Before anyone builds a single lesson, we name three things: who the learner is, what they need to be able to do, and the standard they will be measured against. Skip those and the training misses, every time. A prompt is nothing more than a training request for a machine. Name the same kinds of things up front and the answer changes completely.

The framework for doing that is called CRAFT. I did not invent it. It is a well-known prompting method you will see taught in a lot of places. What I add is how a career instructional designer teaches it, and how to use it without the hype. Here are the five parts.

C, Context. What the AI needs to know about your situation. The more it knows, the less it guesses. “I am retiring from the Navy after 20 years as an Electronics Technician, moving into civilian IT with a CompTIA A+ and no civilian work history yet.”

R, Role. Who you want the AI to be. Point it at an expert lens and the answer focuses. “Answer as a civilian hiring manager who screens IT support resumes.” You do not need to inflate it into “the world’s number one expert.” Naming the real role does the work.

A, Action. The one specific thing you want done. One clear verb beats a pile of vague ones. “Draft a cover letter for a help desk technician role.”

F, Format. The shape you want the answer in. This is the single biggest lever most people skip. “Three short paragraphs, under 250 words.”

T, Target and Tone. Who the answer is for and how it should sound. “It is for a civilian recruiter who never served, so keep it plain and confident with no military acronyms.”

Put those together and the vague “write me a cover letter” becomes a request the AI can actually fill. Same tool. Same thirty seconds of typing. A completely different answer, one that sounds like you and fits the job.

You do not need all five every time. Even two or three, usually Context and Format, will transform what you get back. And when the stakes are high, add one guardrail after CRAFT: “Do not make anything up. If you are not sure, say so.” That keeps the answer honest.

You do not need to be technical to get real work out of AI. You need to know how to ask, and CRAFT is how you ask.

I am building a short, free course that walks through CRAFT step by step for people leaving the service. Start with The Foundation, our free starter course, at https://rhynowerks.ai.

Authored and reviewed by David M. Moore for Rhynowerks LLC, and produced with AI assistance.