Prompting Like a Professional: The CRAFT Methodology

How to turn a vague request into work you can actually use — in five questions, every time. A complete walkthrough of the CRAFT prompt scaffold: Context, Role, Action, Format, Tone — with end-to-end examples for engineering, legal, project management, marketing, and finance.

You have a draft email open. Or a status update. Or a technical specification that needs to be summarized for your steering committee. You toggle to Copilot, ChatGPT, or Claude. You type something like “summarize this” or “draft an email about X.” You hit enter.

You read the output. It’s fluent. It’s plausible. It’s also generic — and now you have to spend ten minutes editing it into something that sounds like you, says what you meant, and goes to the right audience.

You are not bad at AI. AI is not bad at the work. The problem is the prompt.

This is the gap CRAFT is built to close. Not by making your prompts longer — by removing the ambiguity that makes AI’s first answer generic in the first place. Five letters. Five questions you answer before you hit enter. Once it becomes a habit, the difference between a wasted ten minutes and a useful first draft is one minute of upfront thinking.

This post walks you through the entire CRAFT methodology — what each letter is, why each one matters, and how the five fit together. We’ll work through one engineering example end-to-end, then show how the same scaffold lands in project management, legal, marketing, and finance work. By the end, you’ll have a structure you can use on your very next prompt.

Why Most Prompts Fail

Most prompts fail in the same way: they assume the AI knows things it doesn’t.

It doesn’t know who your audience is. It doesn’t know what got decided in the meeting last Tuesday. It doesn’t know that your steering committee is non-technical and prefers three short paragraphs over a bulleted list. It doesn’t know your firm has a house style that avoids exclamation points and the word “leverage.” It doesn’t know the constraint that this draft can only be 200 words because the executive reads on a phone.

When you don’t tell it any of that, AI fills the gaps with its defaults. Its defaults are generic — fluent, structured, plausible, but generic. You then spend the next ten minutes editing the genericness out.

CRAFT is the discipline of saying out loud what you would have left implicit. It is not theater. It is not over-engineering. It is the same upfront clarity you would give to a junior engineer or a new hire — written down, in a structure that holds together.

Architects don’t start with detailed blueprints. They start with rough sketches. Each sketch is a step closer. Your first prompt is the rough sketch. The skill is the second one — refining what came back rather than starting over. CRAFT is what makes the first sketch worth refining.

CRAFT — Five Letters, Five Questions

| Letter | Question to answer | | — | — | | C — Context | What does the AI need to know about the situation? | | R — Role | What perspective should the AI take? | | A — Action | What exactly should the AI do? | | F — Format | How should the output be structured? | | T — Tone | What voice and register should it use? |

Five questions. You answer them in your head — or, when you’re learning, on a notecard — and then you write the prompt. The output you get back is not just better. It is recognizably yours, even on the first turn.

C — Context

Context is the single highest-leverage letter in CRAFT. If you skip nothing else, do not skip this one.

Context tells the AI three things:

1. What the situation is — the project, the meeting, the document, the workflow you are inside of.
2. Who the audience is — the steering committee, the client, the engineering team, your boss, a regulator.
3. What’s already been decided — the constraints, the prior decisions, the things that are settled and don’t need re-litigating.

When you skip context, the AI defaults to a generic professional context. Generic professional voice is recognizably empty. It has the cadence of writing without the texture of writing. Once you’ve seen enough generic AI output, you can spot it from across a room — and so can anyone you send it to.

A context-free prompt:

Summarize this technical specification.

A prompt with context:

This is the technical specification draft for a hydrogen-based direct reduction iron plant we are designing for an internal stress-test exercise. I’m preparing a 200-word summary for our client steering committee, who are not engineers but who fund the project. The decision in front of them next week is whether to approve the next funding tranche.

Same task. The first one will produce a serviceable but generic summary. The second one will produce a summary that knows the audience is non-engineers, knows the document is at draft stage, and knows there’s a decision attached.

Context is what AI needs to do its job. Not what AI needs to be impressed by you. The discipline of writing context is also the discipline of clarifying — for yourself — what you actually want.

R — Role

Role tells the AI whose voice to take. It is the second-highest leverage letter.

Without a role, the AI takes the average of every voice it has ever read on the internet. With a role, it takes a specific perspective — and the output sharpens accordingly.

Act as a senior process engineer with twenty years of experience on hydrogen-DRI plants.

Act as the project manager preparing for a steering committee that has read the last update.

Act as a contract attorney reviewing for risk, not for elegance.

The role is not a costume. The role is a register. A senior process engineer writes more bluntly than a junior one — fewer hedges, more domain-specific vocabulary, more comfort with stating risks plainly. When you name the role, the AI calibrates the register.

A useful default if you don’t know the role:

Act as an experienced [your job title] at a [type of firm].

That’s almost always good enough for the first turn.

A — Action

Action is the verb. It is what you actually want to happen.

Most “prompts that don’t work” fail here because the verb is vague. “Help me with…” is not a verb. “Look at…” is not a verb. “Tell me about…” is not a verb.

Useful verbs are specific:

• Draft — produce a first version. Implies the human will edit.
• Summarize — compress while keeping fidelity.
• Compare — structure the difference between two things.
• Critique — surface what’s wrong, missing, or weak.
• Identify — list specific items meeting specific criteria.
• Extract — pull specific elements out of a longer document.
• Translate — render the same content in a different register or for a different audience.
• Format — restructure existing content into a specific shape.

What’s not fine: a verb that hides another verb inside it. “Help me with my email” has at least three verbs hiding inside it. Pick one.

F — Format

Format is what the recipient sees first.

The same content, formatted differently, is a different artifact. A bulleted list of risks is a different document than a three-paragraph narrative summary of the same risks, even if the underlying analysis is identical.

Good format specifications include:

• Length — a 200-word executive summary versus a 2-page brief.
• Structure — bullet list, numbered list, three-column table, sectioned document with headers.
• Visible elements — a “Decision needed” line at the end, a summary at the top, a table of changes.
• Stylistic choices — Oxford commas, bullet style, header level.

Specific:

Format the output as three short paragraphs labeled Progress, Risk, Decision Needed. End with one sentence: “What I need from you this week is ___.”

Vague:

Make it look like a status update.

The first one drops cleanly into a steering committee deck. The second one produces something approximately status-update-shaped that you’ll spend five minutes restructuring.

T — Tone

Tone is voice and register. It is the texture of the writing.

Tone choices to be specific about:

• Formality — formal, semi-formal, conversational, casual.
• Hedging — confident and direct versus measured and cautious.
• Technical density — assume the reader knows the domain, or explain as you go.
• House style — your firm’s actual house style, named explicitly.

Useful tone specifications:

Tone: factual, technical, no marketing language.

Tone: confident, plain English, no hedging, no closing platitudes.

Tone: collegial, precise, the way a senior engineer would speak to a peer they respect.

The interesting thing about tone is that it carries implications about the rest of the prompt. “Confident, no hedging” implies a reader who can take direct feedback. Tone is how you tell the AI who the reader actually is — beyond what Context alone can do.

CRAFT Walked End-to-end — The Engineering Example

Here’s the same prompt, fully built in CRAFT:

Context: This is the technical specification draft for a hydrogen-based direct reduction iron plant we are designing for an internal stress-test exercise. I’m preparing a summary for our client steering committee. The committee members are not engineers, but they fund the project. The decision in front of them next week is whether to approve the next funding tranche. Last week’s update covered scope and budget; this week the open question is technical risk.

Role: Act as a senior process engineer with twenty years of experience on hydrogen-DRI plants, translating for an executive audience.

Action: Summarize the attached specification draft, focusing on the elements that drive the funding-tranche decision.

Format: Four short paragraphs of roughly 50 words each, labeled: Scope, Key Technical Choices, Current Risks, Decisions Needed. Total length 200 words.

Tone: Formal, plain-English where possible, no marketing language. No hedges where the underlying engineering is settled. Hedges only where they reflect actual uncertainty.

What you get back is a summary that knows it is for non-engineers, knows the funding-tranche decision is the focal point, and knows the audience is paying attention to risk this week. It will be 80% of the way there on the first turn.

Compare to: Summarize this spec. You get something fluent and generic. You spend ten minutes editing in the audience-awareness, the framing, the labeled paragraphs, the tone calibration. CRAFT moves that ten minutes upfront — and turns it into one minute of structured thinking.

CRAFT in Other Domains

Project management — weekly status synthesis

Context: Weekly status update for the project steering committee. Non-engineers who fund the work. Last week we asked them to approve a vendor change; they approved it. This week the open question is whether the new vendor’s onboarding timeline puts us behind on the integration phase.

Role: Act as the project manager.

Action: Synthesize the bullet inputs into a status update. Resolve any apparent contradictions between sub-team inputs by flagging them as decisions-needed rather than smoothing them over.

Format: Three short paragraphs labeled Progress, Risk, Decision Needed. End with one sentence: “What I need from you this week is ___.”

Tone: Confident, plain English, no PM jargon, no hedging where the data is clear.

Legal — contract review pre-pass

Context: Draft master services agreement from a new vendor. Engagement value approximately $500K over twelve months. Our negotiation playbook prioritizes liability caps, IP assignment, and data protection clauses.

Role: Act as a contract attorney reviewing for risk, not for elegance.

Action: Identify clauses that diverge from our playbook positions on liability, IP, and data protection. For each, summarize the divergence, the business risk, and the suggested redline.

Format: Three-column table — Clause Reference | Divergence and Risk | Suggested Redline. Sort by severity, most material first. Limit to top eight items.

Tone: Neutral, factual. No softening language. State the risk plainly.

Marketing — landing page first draft

Context: Launching a new product page for our enterprise platform. Audience is heads of operations at mid-market manufacturing firms. The page needs to convert a first-time visitor into a demo request. Our brand voice avoids superlatives. Our differentiator is measurable time savings on operational workflows.

Role: Act as a senior copywriter who has written for B2B SaaS companies that sell to operations leaders.

Action: Draft a landing page first version with hero, three feature blocks, social proof section, and primary CTA.

Tone: Warm, specific, calmly confident. No “transform your business” language. Active voice.

Finance — variance analysis narrative

Context: Q3 close just finished. Revenue came in at $42.3M against a budget of $45.0M — a $2.7M unfavorable variance. Concentrated in the North America segment, which underperformed by $3.1M, partially offset by $0.4M favorable in EMEA. Three drivers: contract renewal slippage, FX tailwind in EMEA, and a $0.6M deferred revenue true-up from Q2. The CFO will brief the audit committee next week.

Role: Act as the FP&A senior analyst preparing the variance narrative for the CFO’s audit committee briefing.

Action: Draft a variance narrative walking through the unfavorable revenue variance, the three drivers, and the implications for Q4 forecast.

Format: Four sections — Headline, Driver Analysis (three labeled paragraphs), Forecast Implication, Asks. Total length 350–400 words.

Tone: Measured, factual, audit-committee appropriate. Conservative on forward-looking statements.

Five different professions. Same scaffold. The output in each case lands close enough to ready that the human edit becomes a refinement, not a rewrite.

The Refinement Loop — What to do With CRAFT’s Output

CRAFT gets you a strong first draft. It does not get you a final draft. The skill that separates capable users from fluent ones is what happens after CRAFT — the second turn.

Three refinement turns are useful most of the time:

Tighter scope. When the draft is right but too long or covers too much.

Cut to 150 words. Drop the “risks” paragraph; add a one-sentence “next decision” line at the end.

Different audience. When the draft is right but written for the wrong reader.

Now rewrite for the engineering team — assume they know the technical context. Same length, more technical density, less hedging.

Surface a specific concern. When the draft is right but doesn’t say the thing you most need to know.

What is the most likely point of failure in this spec that the client steering committee should know about? One paragraph.

The third refinement type is the most underused — and often the most useful. CRAFT plus a “what’s the biggest risk a reviewer would flag” turn is a remarkably effective workflow for surfacing the question you didn’t ask.

Common Mistakes — and How to Recognize Them

Five patterns trip people up most often.

Skipping Context because it feels redundant. Context is the single highest-leverage letter. The thing that feels redundant to you is the thing the AI doesn’t have. Write it anyway.

Choosing a generic Role. “Act as an expert” is a non-role. Pick a specific role with a specific register — the more specific, the better the output.

Burying multiple Actions inside one verb. “Help me with this” hides at least three verbs. Pick one.

Format that’s too vague. “Make it look professional” is not a format. Specify length. Specify structure.

Tone that’s secretly two tones. “Confident but humble” is a contradiction the AI will resolve in favor of one or the other. If you mean “confident on the data, humble on the forward look,” say that.

The pattern across all five mistakes is the same: the AI defaults to genericness when you leave room for it. CRAFT closes that room.

The 80% Rule

CRAFT will not get you to 100%. It will reliably get you to 80% on the first turn. The remaining 20% is what makes the work yours — your house edits, your judgment about what to cut, your knowledge of which sentence will land with this specific reader.

That’s the right division of labor. AI does the tax — research, formatting, drafting, summarizing. You do the judgment — what matters, what doesn’t, what’s right, what’s wrong. CRAFT is the structure that lets the tax get removed cleanly so judgment can be applied to more decisions.

It is also why “AI will replace your job” misses the point. The judgment is the job. CRAFT just makes the job faster.

What to do Tomorrow Morning

Pick one task you’ll do in the next seven days that involves writing, summarizing, or drafting something. Don’t pick the hardest one — pick a normal Tuesday-afternoon one.

Before you open Copilot or ChatGPT or Claude, write five lines on a notecard:

• C — Context: …
• R — Role: …
• A — Action: …
• F — Format: …
• T — Tone: …

Write the prompt from the notecard. Run it. Read the output.

Then run one refinement turn — tighter scope, different audience, or surface a specific concern.

That’s it. That’s the entire workflow. Once you’ve done it three times it becomes the way you prompt. The notecard goes away. The structure stays.

Final Note — Clear Intention Creates Clear Collaboration

There is one phrase we leave with every participant in every WorkSHIFT cohort:

Clear intention creates clear collaboration.

CRAFT is the discipline of clear intention, written down. The collaboration that follows — between you and the AI, between you and the colleague reading the output, between you and the deadline you are trying to meet — gets clearer because the intention got clearer first.

The AI is a reflection pool. Bring clear thinking to it, it reflects clear work back. Bring scattered thinking, it reflects scattered work. The pool doesn’t judge. It amplifies what you bring.

Bring CRAFT.

This post draws on the WorkSHIFT methodology and the CRAFT prompt scaffold taught in C1M’s WorkSHIFT training program. The 12-template CRAFT cheat sheet — covering engineering, project management, procurement, operations, legal, marketing, and finance — is included with every WorkSHIFT engagement. To learn more about how WorkSHIFT could land at your firm, visit workshift-training.