5 Ways to Write Better Prompts (And Actually Get the Results You Want from AI)

Most people use AI the same way they use Google — they type a short, vague query and hope for the best. The problem is that AI isn't a search engine. It's more like a highly skilled collaborator who needs proper direction. The better your instructions, the better the output.

This guide breaks down five practical techniques you can apply right now to sharpen your prompts and get results that are actually useful.

The Core Problem: Vague In, Vague Out

Before we get into the techniques, it's worth understanding why bad prompts produce bad results.

AI language models generate responses based on the patterns and context in your input. When you write "write me a caption," the model has almost no signal to work with — it doesn't know your brand, your audience, your tone, your platform, or what you're promoting. So it gives you a generic answer, because that's all it can do with generic input.

Here are the most common vague prompts people use:

• "Write me a caption" — for what? what brand? what product? what tone?

• "Make it better" — better in what way? shorter? more professional? more emotional?

• "Write a blog post" — about what? for whom? how long? what's the goal?

  
  

BORING!

Every one of these leaves the AI guessing. And when AI guesses, it defaults to average.

The fix isn't complicated. It just requires you to give the model more to work with. Here are the five ways to do that.

Way 01 — Give AI a Role to Play

The single fastest way to improve your output is to assign AI a specific role before you ask it to do anything.

When you tell a model who it is, you're activating a very specific part of its knowledge. A social media strategist thinks differently from a copywriter. A pediatrician gives different advice from a nutritionist. The role shapes the vocabulary, the priorities, and the tone of every response.

The pattern:

Example without a role:

Example with a role:

The second prompt will consistently outperform the first. Not because the task changed — but because the model now has a lens through which to interpret the task.

When to use this: Any time you're creating content, getting advice, or asking for analysis. Role-setting is especially powerful for tone-sensitive tasks like marketing copy, customer emails, or educational content.

Way 02 — Specify the Format

AI will structure its output however it sees fit — unless you tell it otherwise. And the default structure is rarely what you actually need.

If you ask for "a summary," you might get five paragraphs when you wanted three bullet points. If you ask for "ideas," you might get a long paragraph when you needed a numbered list you could scan quickly.

Specifying format isn't about being controlling — it's about reducing the gap between what you need and what you get.

Format options to specify:

• Bullet points or numbered lists

• A table with specific columns

• A 3-sentence summary

• A single paragraph, no more than 80 words

• Headers and subheaders

• A script with speaker labels

• JSON or structured data

• 
  

Example:

That prompt doesn't just ask for a summary — it tells the model exactly how to structure the output, which saves you time editing it afterwards.

Pro tip: You can also specify what not to include. "No hashtags. No emojis. No filler phrases like 'In today's fast-paced world.'" These negative constraints clean up the output significantly.

Way 03 — Add a Constraint

Constraints feel counterintuitive. Why would limiting the AI make it better? Because boundaries force specificity — and specificity produces sharper output.

Think of it like a brief. A designer with a fully open brief will give you something generic. A designer told "it needs to feel trustworthy, target 50+ year olds, and fit in a single A5 card" will give you something focused.

Common constraints to add:

• Word or character count: "under 60 words"

• Audience: "for someone who has never used AI before"

• Tone: "professional but not stiff — like a knowledgeable friend"

• Platform: "optimised for LinkedIn, not Twitter"

• Exclusions: "no technical jargon, no acronyms"

• Reading level: "write at a Grade 8 reading level"

• 
  

Example:

Constraints stack. The more relevant ones you add, the narrower the target, and the more consistently the model hits it.

Way 04 — Use Few-Shot Examples

This is one of the most underused techniques, and it's remarkably effective.

"Few-shot prompting" means giving the model one or two examples of the output you want before asking it to produce its own version. Instead of describing what you want in abstract terms, you show it.

This works because AI models are pattern matchers. When you show them an example of your writing style, your brand voice, or your preferred structure, they learn to mirror it almost instantly.

The pattern:

Another use case — tone matching:

And for structured outputs:

You don't need many examples — one or two is usually enough. The model picks up the pattern fast. This technique is especially useful when you've spent time developing a brand voice and don't want AI to override it with something generic.

Way 05 — Refine, Don't Restart

Most people, when they get a bad response from AI, do one of two things: they accept the mediocre output, or they start the whole prompt from scratch. Both are mistakes.

The right move is to treat the first response as a first draft — and iterate on it using targeted feedback.

This mirrors how you'd work with a human collaborator. You wouldn't fire a copywriter after their first draft and hire someone new. You'd give them notes. The same logic applies here.

What makes iteration effective:

• Be specific about what to change: "The third paragraph is too formal — make it more conversational."

• Tell it what to keep: "Keep the structure, but rewrite the opening sentence — it's too generic."

• Give it a direction: "Make this 30% shorter without losing the key points."

• Reference the original: "Go back to your second version — that tone was closer to what I need."

  
  

What doesn't work:

• "Make it better" (better how?)

• "Try again" (what should change?)

• Starting a new chat and rewriting the prompt from scratch (you lose all the context you've built up)

  
  

One of AI's strengths is that it holds the full conversation in context. Use that. Build on what you've already generated rather than abandoning it.

Putting It All Together: The 5-Part Prompt Formula

Once you've internalised these five techniques, they combine naturally into a single framework you can apply to almost any prompt. Think of it as the RCTFC formula:

Example using all five:

You won't always need all five elements — sometimes two or three is enough. But having the full framework in your head means you'll never stare at a blank prompt box again.

Final Thought

Writing better prompts isn't a technical skill — it's a communication skill. You're learning to articulate your needs more clearly. The more you practise, the faster and more naturally it comes.

Save the RCTFC formula. Use it on your next prompt. Then iterate.

That's it. Now get cracking!