How to Write AI Prompts That Actually Get You What You Want

Most people's first AI prompt looks something like "write me a marketing email." The result is generic, and the temptation is to blame the tool.

The tool usually isn't the problem. The prompt is missing information the AI needs to do the job well.

Good news: you don't need to learn special syntax or memorize magic phrases. You need a repeatable structure. Once you have one, it works the same whether you're using ChatGPT, Claude, Microsoft Copilot, or Google Gemini. These tools differ under the hood, but they all respond to the same basic inputs: who, what situation, what task, and what output.

The RCTF Framework

Here's a simple structure you can reuse for almost anything: Role, Context, Task, Format.

• Role — Who should the AI act as? A role focuses its tone, vocabulary, and priorities.
• Context — What's the background? Include what the AI can't guess: your industry, your audience, constraints, prior attempts.
• Task — What exactly do you want done? Be specific about the action, not just the topic.
• Format — How should the output be structured? Length, tone, sections, bullet points versus paragraphs.

You don't need labeled headers in the prompt itself, though it doesn't hurt. What matters is that all four pieces show up somewhere.

Before and After: A Marketing Email

Weak prompt:

Write a marketing email about our new service.

This will produce something. It just won't sound like your business, and you'll spend more time editing than if you'd written it yourself.

Strong prompt, using RCTF:

You are a marketing copywriter for a small IT support company (Role). We just launched a new service: 24/7 help desk support for small businesses that previously only had business-hours support. Our audience is small business owners in Western Wisconsin who are frustrated by slow response times from their current IT provider (Context). Write a promotional email announcing this service and encouraging them to reply for a free consultation (Task). Keep it under 150 words, use a calm and confident tone (not salesy), and end with a single clear call to action (Format).

The difference isn't length for its own sake — it's that the AI now knows who it's writing as, who it's writing to, what problem it's solving, and what "done" looks like. That's the whole trick.

More Worked Examples

Customer Service Email

Weak: "Write a response to an angry customer."

Strong: "You are a customer support lead at a small managed IT services company (Role). A customer emailed saying their internet went down twice this week and they're considering switching providers; they've been a customer for three years and generally happy until now (Context). Write a reply acknowledging the frustration, explaining we're investigating the root cause, and offering a follow-up call this week (Task). Keep it to 3 short paragraphs, warm but professional, no excessive apologizing (Format)."

Internal Policy Draft

Weak: "Write a remote work policy."

Strong: "You are an HR consultant drafting policy for a 12-person professional services firm (Role). We currently have no written remote work policy; employees have been informally working from home 1-2 days a week and leadership wants to formalize this without making it feel restrictive (Context). Draft a one-page remote work policy covering eligibility, expected availability hours, and equipment responsibility (Task). Use plain language, organize with short headers, avoid legal jargon (Format)."

Data Summary

Weak: "Summarize this spreadsheet."

Strong: "You are a business analyst preparing a summary for a busy owner who has 5 minutes to read it (Role). Attached is a spreadsheet of last quarter's website traffic by source, with columns for source, visits, and conversions (Context). Identify the top 3 traffic sources by conversion rate and flag any source that dropped more than 20% month over month (Task). Output as a short bulleted list, no more than 6 bullets, plain numbers not percentages of percentages (Format)."

Notice the pattern: every strong version tells the AI who it's speaking as, what's actually going on, exactly what to produce, and what shape the output should take. None of these examples used special commands or tool-specific tricks — that's the point. RCTF is portable across whichever AI product your team already has access to.

Why This Still Requires a Human

A framework doesn't remove the need for judgment. You still have to decide what the right task is, what context actually matters, and whether the output is accurate before you send it. Think of RCTF as a way to get a much better first draft, not a way to skip review.

This also means the framework rewards practice. The first few times you use it, you'll find yourself leaving out context you didn't realize you needed. That's normal — refine and re-run rather than starting over.

Prompting Skill Is a Culture and Talent Issue

It's tempting to treat prompting as a personal quirk — some people are just "good at it." In practice, it's a skill you can teach and standardize across a team, the same way you'd standardize how people write emails or run meetings.

Organizations that build this into onboarding and everyday habits get more consistent, more useful output from the AI tools they're already paying for. Organizations that leave it to chance end up with a wide gap between their best and worst AI users — and often conclude the tools "don't work" when the real gap is training.

This is exactly the kind of thing we look at in our AI Business Maturity Assessment, which evaluates how ready your team's culture and talent are to get real value from AI, not just whether you have licenses. If you want help building a simple prompting standard for your team, contact us — we're happy to talk through where to start.