Blog / Prompting & LLM Ops

Stop Asking ChatGPT Bad Questions: A Practical Guide to Prompts That Actually Work for Business

Schuyler WhetstoneJul 18, 20259 min read

TL;DR: The difference between useless AI output and genuinely helpful results comes down to one thing: the quality of your prompt. Use the CRAFT framework — Context, Role, Action, Format, Tone — to transform vague questions into precise instructions that deliver business-ready outputs every time.

The Prompting Problem

You've been lied to about AI.

Not by the AI companies — by the internet. Every "10x your productivity with ChatGPT" thread makes it look effortless. Type a sentence, get magic. But when you try it, you get... meh. Generic. Surface-level. The kind of output that makes you wonder why everyone's so excited about this technology.

Here's the truth: AI mirrors the specificity of your input. Vague question equals vague answer. Every time.

The good news? Fixing this doesn't require a computer science degree. It requires understanding one simple framework and developing a few habits that will transform your AI interactions from frustrating to genuinely useful.

Why Your Current Prompts Don't Work

Let's diagnose the problem. Here's what most people type into ChatGPT:

"Write me an email to a client."

And here's what they get: a generic, corporate-sounding email that could have been written for any client, about any topic, in any industry. It's technically correct and practically useless.

The prompt failed because it gave the AI zero context about:

  • Who the client is
  • What the email is about
  • What tone is appropriate
  • What outcome you want
  • What your relationship looks like

The AI isn't stupid — it's operating with insufficient information. It's like asking a stranger to "write a letter" without telling them who it's for or why.

The CRAFT Framework

CRAFT stands for Context, Role, Action, Format, and Tone. Not every prompt needs all five elements, but the more you provide, the better your output.

C — Context

Context is the background information the AI needs to understand your situation. This is where most people fail — they assume the AI knows things it doesn't.

Bad: "Write a follow-up email."

Good: "I run a 10-person automation consultancy. We had a discovery call with a potential client last Thursday — a mid-size law firm that's drowning in manual document processing. They seemed interested but haven't responded to my proposal sent Monday."

The context doesn't need to be an essay. A few sentences that establish who you are, who you're talking to, and what's already happened transforms the output.

R — Role

Telling the AI who to be changes how it thinks and responds. Different roles produce different outputs from the same information.

Without role: "How should I price this project?"

With role: "You're a senior pricing strategist for B2B service businesses. You specialize in value-based pricing for technology consulting firms."

Useful roles for business:

  • "You're a direct-response copywriter who specializes in B2B SaaS"
  • "You're an operations consultant who helps small businesses streamline workflows"
  • "You're a CFO advising a bootstrapped startup on budget allocation"
  • "You're an experienced sales manager coaching a junior rep"

A — Action

The action is what you want the AI to actually do. Be specific about the deliverable.

Vague: "Help me with my marketing."

Specific: "Write three subject line options for a cold outreach email targeting operations managers at companies with 50-200 employees who are still using spreadsheets for project management."

The more specific your action, the less the AI has to guess — and the less it guesses, the better the output.

F — Format

Tell the AI exactly how to structure its output. This eliminates the most common complaint: "It gave me a wall of text when I wanted bullet points."

Format specifications include:

  • Structure: Bullet points, numbered list, table, paragraph
  • Length: "3 sentences," "under 200 words," "one page"
  • Sections: "Include an introduction, three main points, and a call to action"
  • Style: "Use headers and subheaders," "include a summary at the top"

Example: "Format as a table with columns for: Feature, Benefit to Client, and Implementation Timeline. Keep each cell to one sentence."

T — Tone

Tone shapes how the output feels. The same information delivered in different tones creates completely different impressions.

Examples:

  • "Write in a conversational, approachable tone — like explaining to a smart friend over coffee"
  • "Professional but warm — this client has been with us for 3 years"
  • "Direct and confident — no hedging language or qualifiers"
  • "Technical but accessible — assume the reader is smart but not a developer"

CRAFT in Action: Before and After

Let's transform a real business prompt:

Before (typical prompt)

"Write a proposal for a client."

Output: A generic, template-style proposal that could be for any service, any client, any industry.

After (CRAFT prompt)

Context: I run AIgency Automata, an automation consultancy. We just had a discovery call with Martinez & Associates, a 25-person law firm. They spend 40+ hours/week on manual document intake, client communication, and billing follow-ups. Their office manager, Sarah, is our main point of contact. Budget is around $5,000-8,000.

Role: You're a senior automation consultant writing a proposal that demonstrates deep understanding of the client's pain points.

Action: Write a project proposal that outlines three automation packages (Basic, Standard, Premium) with specific workflows for their use case.

Format: Structure as: Executive Summary (3 sentences), The Problem (their specific pain points), Proposed Solutions (three tiered packages in a table), Timeline, Investment, and Next Steps. Keep it under 1,000 words.

Tone: Confident and knowledgeable but not salesy. Use their specific terminology ("intake," "billing cycles") to show we listened.

Output: A specific, persuasive proposal that references their exact situation, uses their language, and presents clear options with realistic pricing.

The difference isn't magic. It's information.

The Anti-Prompt: Telling AI What NOT to Do

One of the most underused prompting techniques is negative constraints — telling the AI what to avoid.

This works because AI models have strong default patterns. Unless you explicitly override them, you'll get:

  • Corporate jargon ("leverage," "synergize," "optimize")
  • Hedging language ("it might be worth considering")
  • Unnecessary disclaimers ("as an AI, I should note...")
  • Overly formal tone
  • Generic examples

Add constraints like:

  • "Don't use corporate buzzwords — write like a real person talks"
  • "No hedging — be direct and take a clear position"
  • "Skip the introduction — start with the most important point"
  • "Don't include disclaimers about being an AI"
  • "No generic examples — use specific, realistic scenarios from [industry]"

Constraints are often more powerful than instructions because they eliminate the noise rather than trying to describe the signal.

The Iteration Mindset

Here's what most people get wrong: they treat prompting like a Google search. Type once, get the answer, done.

But prompting is a conversation, not a query. Your first prompt is a draft — expect to refine it.

The 3-Turn Rule

Most prompts reach their best output within three turns:

Turn 1: Your initial CRAFT prompt → Get the first draft

Turn 2: "This is good, but [specific adjustment]. Also, make the tone more [specific] and expand on [section]." → Refined output

Turn 3: "Perfect direction. Now tighten the language — cut any sentence that doesn't add new information. Replace the example in section 2 with something about [specific scenario]." → Final output

Three turns. Two minutes. Output that would have taken you 30 minutes to write from scratch.

When to Start Over vs. Iterate

Iterate when: The structure is right but the content needs adjustment (tone, depth, specificity).

Start over when: The AI went in a completely wrong direction. This usually means your initial prompt was too vague — add more context and try again rather than trying to course-correct.

Building Your Prompt Library

The biggest productivity hack isn't a better AI model — it's saving your best prompts.

Think about it: 80% of your AI interactions fall into repeating categories. Weekly reports. Client emails. Social media posts. Meeting summaries. Proposal sections.

Build a library:

  1. When a prompt produces great output, save it — Don't just save the output. Save the prompt that generated it.
  2. Template the variables — Replace specific details with placeholders: [CLIENT_NAME], [PROJECT_TYPE], [DEADLINE]
  3. Organize by category — Emails, Proposals, Content, Analysis, Internal Comms
  4. Iterate the templates — Each time you use a saved prompt, note what worked and what needed adjustment. Update the template.

After a month, you'll have a personalized prompt library that produces consistent, high-quality output for your most common tasks. This is the real competitive advantage — not knowing secret prompts, but having proven templates tuned to your specific needs.

Platform-Specific Tips

ChatGPT

ChatGPT is forgiving with unstructured prompts — it's good at inferring intent from casual language. But this can also make it lazy about following precise instructions. Be explicit about format and constraints.

Pro tip: Use Custom Instructions to set your default context (your role, company, preferred tone) so every conversation starts from a better baseline.

Claude

Claude tends to follow detailed instructions more precisely than ChatGPT. It excels with long, structured prompts and is better at maintaining consistency across multi-part outputs.

Pro tip: Claude responds exceptionally well to examples. Show it one example of the output you want, and it'll match the pattern consistently.

Gemini

Gemini integrates well with Google Workspace, making it ideal for tasks that involve your existing documents and data.

Pro tip: When using Gemini with Google Docs or Sheets, reference specific documents by name for better context.

Common Prompting Mistakes (and Fixes)

MistakeWhy It FailsFix
Too vagueAI has to guess everythingAdd context and specifics
Too longKey instructions get buriedFront-load the most important parts
No format specifiedOutput structure is randomAlways specify desired format
Asking for "the best""Best" is subjective and undefinedDefine your criteria explicitly
One-and-done mentalityMissing the iteration benefitPlan for 2-3 refinement turns
No negative constraintsAI defaults to generic patternsTell it what to avoid

The 5-Minute Prompting Habit

Before typing any prompt, spend 30 seconds answering these questions:

  1. What do I actually need? (Not "help with marketing" but "three email subject lines for cold outreach")
  2. What context is missing? (Industry, audience, relationship, history)
  3. What should this look like? (Format, length, structure)
  4. What should this NOT include? (Jargon, disclaimers, generic advice)

Thirty seconds of thinking before prompting saves five minutes of re-prompting after.


Related Reading

CTA (sonnet, for better prompts):

If every answer feels like beige and thin,

The fault is not the model—start upstream;

Give context first, the role it must step in,

Name what you want, and how it should be seen.

Constrain the noise; forbid the jargon cloud;

Ask for a table, bullets, tight and clean;

Then iterate—make drafts say it out loud

Until the output matches what you mean.

If you want a prompt library that converts,

Send one real task you do each week on repeat;

I’ll craft a CRAFT template that doesn’t flirt—

It lands the point, then makes the work feel neat.

Reply with audience, goal, and your “don’t do,”

And I’ll hand back a prompt that follows through.


Frequently Asked Questions

Why do my ChatGPT outputs feel generic?

Because your prompts are generic. AI mirrors the specificity of your input. Vague question equals vague answer. Add context about your business, audience, and desired outcome.

What's the best prompting framework for business?

CRAFT: Context (background info), Role (who the AI should be), Action (what to do), Format (structure of the output), Tone (voice and style). Not every prompt needs all five, but the more you provide, the better the output.

Should I use the same prompts for ChatGPT and Claude?

The same frameworks work, but Claude tends to follow detailed instructions more precisely, while ChatGPT is better at inferring intent from less structured prompts. Adjust based on which model you're using.

How long should a prompt be?

As long as it needs to be. A 3-sentence prompt that includes context, action, and format will outperform a 1-sentence prompt every time. Don't be afraid of detailed prompts.


Keep Reading