How to Write Better AI Prompts (With Examples)

Most people use AI wrong. Not because they're bad at it. Because nobody taught them the actual framework.

You ask ChatGPT a vague question. It gives you a vague answer. You think AI is "fine but not amazing." In reality, AI is incredible — you're just not telling it what you want.

Here's the framework that changes that.

The CTFC Framework

Every prompt you write should have four things. When you miss one, your outputs go sideways.

C — Context T — Task F — Format C — Constraints

That's it. Not complicated. But it works everywhere.

C — Context

Context is why you're writing this. Who's the audience? What's the situation? What's the tone?

Bad: "Write about AI in marketing."

Good: "I'm writing a LinkedIn post for marketing managers about using AI in their daily work. They're skeptical but open-minded. I want them to feel like AI is achievable, not intimidating. Audience: UK-based B2B SaaS marketing leads."

Context does two things:

It tells the AI what it's working with
It prevents generic garbage in the output

When you give context, you get specificity. When you don't, you get templates.

T — Task

The task is exactly what you want the AI to do. "Write this." "Summarize that." "Give me alternatives." "Explain why."

Bad: "I need help with an email."

Good: "Write a follow-up email to a prospect who didn't respond to my first email 2 weeks ago. The goal is to re-engage without sounding desperate. Keep it to 3 short paragraphs."

The difference: "help with an email" could mean anything. "Write a 3-paragraph follow-up email" is a task the AI can actually complete.

F — Format

Format is the shape of the output. Bullet list? Paragraph? Table? Short? Long? Simple language or technical?

Bad: "Give me ideas for content."

Good: "Give me 10 blog post ideas for a productivity app. Format as a list. Each idea: title, 1-sentence description, and the main audience. Keep descriptions under 15 words."

Format matters because:

It prevents walls of text when you want bullets
It makes the output immediately usable
It stops the AI from overthinking

C — Constraints

Constraints are the rules. What's off-limits? What should you avoid? Any limits on length, jargon, or topic?

Bad: "Write a bio."

Good: "Write a 100-word professional bio for my LinkedIn. I'm a product manager at a SaaS startup. I've been in tech for 8 years. Tone: confident but not arrogant. Don't mention specific company names. Assume the reader knows nothing about me."

Constraints prevent:

Rambling (it knows when to stop)
Off-brand tone (it knows what's not allowed)
Bad assumptions (it doesn't make things up)

Putting It Together: Real Examples

Example 1: Email Draft

Context: I'm writing a customer who's about to churn. They signed up 3 months ago but have been inactive for 2 weeks.

Task: Write an email that acknowledges they might be losing interest, offers help, and gives them one clear action to take.

Format: Keep it short. 2 paragraphs max. Conversational tone.

Constraints: Don't be pushy. Don't offer a discount. Don't mention specific features.

Prompt: "I have a SaaS customer who signed up 3 months ago but has been inactive for 2 weeks. They might be about to churn. Write an email that's honest about what I've noticed, offers genuine help, and gives them one clear thing to do next. Keep it to 2 paragraphs. Tone: conversational, not corporate. Don't offer discounts or mention specific features."

What you get: An email that actually works because you told the AI exactly what you wanted and why.

Example 2: Content Brief

Context: I'm writing a LinkedIn post for my audience of freelancers who are considering AI tools.

Task: Give me the outline for a post about using Claude for client proposals.

Format: Outline with main points and talking points under each.

Constraints: Assume they've heard of ChatGPT but maybe not Claude. Don't use jargon.

Prompt: "I'm writing a LinkedIn post for freelancers who are considering AI tools. Many have heard of ChatGPT but not Claude. Write me an outline for a post about using Claude to write better client proposals. Include main points and 2-3 talking points under each. Assume the audience is technically capable but new to AI. No jargon."

What you get: An outline you can actually write from because the AI knows your audience and your goal.

Example 3: Code Review

Context: I'm a junior developer asking for feedback on my first API function.

Task: Review this code for bugs, efficiency, and readability.

Format: Keep it organized. Start with overall impression, then specific issues, then suggestions.

Constraints: Explain why something is better, not just what. Assume I'm learning.

Prompt: "Review this code. I'm a junior developer learning to build APIs. [paste code]. Look for: bugs, performance issues, readability. Format your review as: 1) Overall impression, 2) Specific issues with explanations, 3) 3-5 concrete suggestions for improvement. Explain the why behind each suggestion."

What you get: Feedback that teaches you, not just criticism.

Example 4: Research Summary

Context: I need to understand the current state of AI in HR.

Task: Summarize what's changing, what's hyped vs real, and what actually matters.

Format: 3-point summary with evidence.

Constraints: Focus on practical applications, not theory.

Prompt: "[Paste 3 articles about AI in HR]. Summarize these into: 1) What's actually changing in HR, 2) What hype you can ignore, 3) What companies should focus on now. Keep it practical, not theoretical."

What you get: A distilled summary instead of reading 3 articles yourself.

The Common Mistakes

Mistake 1: Assuming the AI Knows Your Context

Bad: "Should I use ChatGPT or Claude?"

Claude doesn't know:

What you use it for
Your budget
Your workflow
What matters to you

Good: "I write a lot of long-form blog posts and product documentation. I sometimes need to summarize PDFs. I'm on a free tier. Should I use ChatGPT or Claude for this? Walk me through the trade-offs."

Mistake 2: Forgetting to Say Exactly What You Want

Bad: "Write a marketing email."

Good could mean:

Cold outreach?
Customer win-back?
Launch announcement?
Newsletter?

Good: "Write a cold outreach email to a potential client. They're the CMO of a mid-size SaaS company. I'm selling a data analytics tool. First email, so assume they don't know me. Goal: Get a 15-minute call. Keep it to 3 short paragraphs."

Mistake 3: Not Specifying the Format

Bad: "Give me social media ideas."

Does that mean:

One idea or ten?
Just titles or full posts?
For which platforms?
How long?

Good: "Give me 5 LinkedIn post ideas about AI productivity. For each: title, 2-sentence hook, estimated engagement (why this will work). Format as a numbered list."

The Template (Copy This)

Use this structure for literally any prompt:

Context: [Why are you writing this? Who's the audience? What's the situation?]

Task: [Exactly what should the AI do?]

Format: [How should the output look?]

Constraints: [What should it avoid? Any limits?]

Prompt: [Your actual message to the AI, combining all of the above]

One More Thing: Iteration

Even with perfect prompts, sometimes the first output isn't perfect. And that's fine.

First prompt: "Write a LinkedIn post about AI and productivity."

Output: [Something good, but a bit too corporate]

Second prompt: "Good, but make it more conversational. Add a specific example of how I use AI in my actual workflow. Remove jargon."

Output: [Better]

Third prompt: "Better. Now add a call-to-action at the end asking people what AI tool they're using and why."

Output: [Perfect]

This is how you actually use AI. Not one perfect prompt. Prompt + iterate + refine.

The Real Skill

The real skill isn't knowing the CTFC framework. The real skill is knowing exactly what you want before you write the prompt.

Most people ask AI bad questions because they haven't thought through what they actually need.

Spend 30 seconds writing down:

Who is this for?
What's the goal?
What should it look like?
What matters most?

Then plug those into your prompt.

Your outputs will improve immediately.

Next Steps

The Cleo AI Starter Guide covers prompting in depth with 45 step-by-step exercises for ChatGPT, Claude, and 7 other tools. Includes templates you can copy for emails, brainstorming, research, and more. Get the guide for $9 →