The AI is not the problem.
If you've ever been frustrated by a mediocre output from ChatGPT, Claude, or Gemini — a blog post that felt generic, a summary that missed the point, code that almost worked but not quite — the issue was almost certainly the prompt, not the model.
This isn't a criticism. It's actually good news. Because it means the fix is entirely in your hands.
In this article, I'm going to show you exactly how to write prompts that get dramatically better results — with real before-and-after examples for each technique. No theory. Just practical frameworks you can use today.
Why Most Prompts Underperform
AI models are trained to be helpful, and "helpful" in the absence of specific instructions means "average." When you give a vague prompt, the model fills in the blanks with whatever seems most commonly useful — which is usually the most generic, middle-of-the-road version of what you asked for.
That's not failure. That's the model doing its job with incomplete information.
The moment you give it more to work with — context, constraints, tone, format, audience — the output transforms. The model was always capable of producing something great. It just needed you to tell it what "great" looked like.
The CRAFT Framework for Writing Better Prompts
After testing hundreds of prompts across multiple models, the best-performing ones consistently share five elements. I call it the CRAFT framework:
- C — Context
- R — Role
- A — Audience
- F — Format
- T — Tone
You don't need all five in every prompt. But the more of these you include, the better your output will be. Let me show you what each one does.
C — Context: Tell It What It Needs to Know
Context is the background information that shapes the response. Without it, the model guesses. With it, the model tailors.
Without context: "Write an email to my client about the project delay."
With context: "Write an email to a client who paid $8,000 for a website redesign. The project is two weeks behind schedule because our main developer got sick. The client is generally patient but values transparency. This is our third project together."
Same task. Completely different email. The second one will sound human, specific, and appropriately professional — because the model has what it needs to write that way.
R — Role: Give the AI a Character to Play
Assigning a role activates a specific "mode" in the model. It shifts how the AI frames its knowledge and chooses its language.
Without role: "Explain machine learning."
With role: "You are a university professor who specializes in explaining complex technical concepts to non-technical business executives. Explain machine learning in a way that's accurate but entirely free of jargon."
The role doesn't just change the vocabulary — it changes the structure, the analogies, the depth, and the assumptions the model makes about what you already know.
A — Audience: Who Is This Actually For?
AI writes differently for a 22-year-old freelancer than for a 50-year-old CFO. Specifying the audience is one of the highest-leverage things you can do in a prompt.
Without audience: "Write a post about the benefits of using AI for small businesses."
With audience: "Write a LinkedIn post for small business owners aged 35–55 who are skeptical about AI and worried it's too complicated or expensive. Speak to their specific concerns before introducing the benefits."
The second prompt produces content that actually connects with real people — because it knows exactly who those people are and what they're thinking.
F — Format: Define What You Want Back
If you don't specify a format, the model picks one. Sometimes it guesses correctly. Often it doesn't. Specifying format removes that uncertainty entirely.
Without format: "Summarize this article."
With format: "Summarize this article in exactly three bullet points. Each bullet should be one sentence. Focus only on the practical takeaways, not background information."
Format instructions can cover: length (word count, number of paragraphs, number of bullets), structure (headers, numbered list, table, prose), what to include, and what to leave out.
T — Tone: Name the Feeling You're Going For
Tone is the difference between content people actually read and content they skim past. Most people forget to specify it.
Without tone: "Write a product description for my AI course."
With tone: "Write a product description for my AI course. The tone should be confident and direct — like a mentor who's been where the reader is and knows exactly how to help. No hype, no empty promises. Conversational but professional."
Useful tone descriptors: confident, warm, authoritative, conversational, playful, direct, measured, energetic, calm, empathetic. Pick one or two that match your brand and include them in every prompt where voice matters.
Before and After: Full Prompt Examples
Let me put the full CRAFT framework into action with three complete examples.
Example 1: Blog Post Intro
Before: "Write an intro for a blog post about remote work."
After: "You are a productivity writer for a business audience. Write a 100-word blog post introduction for an article titled 'Why Remote Work Is Still Misunderstood in 2026.' The audience is managers and team leads at mid-size companies who are skeptical about remote work's long-term viability. Open with a surprising statistic or counterintuitive claim. Tone: direct and slightly provocative."
Example 2: Cold Email
Before: "Write a cold email to a potential client."
After: "You are a senior B2B sales consultant. Write a cold email to the head of marketing at a mid-size e-commerce company. I sell an AI-powered analytics tool that reduces time spent on reporting by 60%. The email should be under 120 words, feel personal rather than templated, have no subject line (I'll write that separately), and end with one soft, low-commitment CTA — asking for a 20-minute call, not a demo."
Example 3: Code Review
Before: "Review my code."
After: "You are a senior Python developer reviewing code for a junior developer. Review the following function for: logic errors, edge cases that aren't handled, readability, and performance issues. For each issue you find, explain why it's a problem and suggest a specific fix. Don't rewrite the whole function — just point out the issues clearly."
The One Habit That Changes Everything
All of these techniques are useful. But the single habit that will improve your prompting faster than anything else is this: read your output critically and iterate.
The first response from an AI is rarely the best one. But it tells you something. If the tone is off, adjust the tone instruction. If the format isn't right, specify it more precisely. If it's too long, add a length constraint. If it missed the point, give more context.
Good prompting is a dialogue, not a one-shot task. Treat the first output as a draft, not a final answer.
Over time, you'll develop an intuition for what each model responds to — and your first-pass prompts will get dramatically better as a result.
When You Don't Want to Think About It
There's one more option, and it's the most honest one I can give you.
Sometimes you just need to get something done. You don't have time to craft the perfect prompt from scratch. You need a good output, quickly, without the back-and-forth.
That's exactly what a well-built prompt library is for. Instead of applying the CRAFT framework from zero every time, you start from a prompt that's already structured — already has the role, audience, format, and tone built in — and you just fill in the specific variable.
That's why I put together 1,495 AI Prompts That Actually Work — a complete library of copy-paste-ready prompts, tested across ChatGPT, Claude, and Gemini, organized by use case. Each one already has the structural thinking done. You show up, customize one variable, and get a great output in under a minute.
It's the difference between building a house from raw lumber and starting from a solid blueprint.
Summary: What to Do Right Now
- Pick one task you use AI for regularly
- Rewrite your usual prompt using the CRAFT framework — add context, a role, an audience, a format, and a tone
- Compare the output to what you usually get
- Adjust and iterate until the output is genuinely something you'd use
Do that with three or four tasks, and you'll have a small personal library of prompts that work. Do it consistently, and you'll see the compounding effect that serious AI users talk about — where each week, AI is saving you more time than the week before.
→ Skip the trial and error — get 1,495 tested, CRAFT-structured prompts ready to use: 1,495 AI Prompts That Actually Work
Related articles:
.webp)
){kind=link}
0 Comments