How to Write Better AI Prompts: A Simple Framework for Powerful Results

Miguel Monzones
Sep 9
5 min read

Generative AI: Your New Best Friend

Ever typed a prompt into ChatGPT and thought, “Hang on, that’s not quite right”? You’re not alone. The trick isn’t that the AI is faulty; it’s that prompting is a skill. And like any skill worth having, practice helps.

Lately, I’ve been knee-deep in AI. Courses, podcasts, articles, all of it, sometimes even dreaming about algorithms over my morning coffee. There’s an avalanche of information out there, and it can be overwhelming.

So let me cut through the noise. Instead of dumping everything I’ve learned, we’ll focus on what matters when you start your AI journey. First, not everything labelled “AI” is truly artificial intelligence. Many tools are simply clever programming. Real AI learns, adapts, and improves with your input.

Generative AI is where this gets exciting. It is powerful, and the magic shows up when you master your prompts. Think baking. Better ingredients, your prompts, make tastier results.

Enter a simple framework: T–C–R–E–I, Task, Context, References, Evaluate, Iterate. Use it to craft prompts that actually deliver. And remember, prompting is not set it and forget it. It is an ongoing process where every attempt sharpens the next.

Ready to dive in? Let’s build prompts that work.

1) The Prompting Framework (T–C–R–E–I)

Task – be crystal clear about what you want

Specificity is your best friend. If you are vague, the AI will be vague.

Explanation: State the action, format, scope, and focus.
Weak: “Write me an article about fitness.”
Strong: “Write a 600-word beginner’s guide to home workouts. Focus on bodyweight exercises, make it easy to read, and include 3 example routines.”

👉 See the difference? You told the AI what to do, how long to make it, and what to include.

Context – set the stage with background info

Think of this as painting the picture so the AI knows where it is operating.

Explanation: Define audience, purpose, tone, and constraints.
Example: “Write a LinkedIn post about our new eco-friendly water bottle. Target young professionals, emphasize sustainability, and keep the tone casual but inspiring.”

👉 Why it works: the model knows who it is speaking to, what to highlight, and how it should sound.

References – share examples to guide style or content

Inspiration needs fuel. Examples anchor the response.

Explanation: Add sample texts, case studies, or stylistic cues. Two to five references usually hit the sweet spot.
Example: “Write a social caption for a product launch in the style of Nike or Apple. Keep it bold and motivational.”
Sample output: “This isn’t just a bottle. It’s a statement. Designed for tomorrow, carried by you today.”

👉 References help the model mirror the right tone and quality.

Evaluate – review the output critically

Do not take the first draft and run.

Explanation: Check against your brief. Length, tone, accuracy, and usefulness.
Example: You asked for 600 words, got 1,200. You wanted casual, got formal.

👉 If it misses, do not scrap it, refine it.

Iterate – tweak and refine the prompt

Prompting is a loop, not a one-and-done.

Explanation: Adjust instructions based on what you got. Add detail, change tone, reframe the task.
First attempt: “Summarise this article.” → too vague.
Iteration: “Summarise this article in 5 bullet points at a ninth-grade reading level. Keep each point under 20 words.”

👉 Much better. Small tweaks lead to stronger results.

Remember ABI: Always Be Iterating

The more you refine, the closer you get to what you need.

2) Applying the Framework

Let’s walk it in practice: brainstorming a new sneaker line.

Initial prompt: “Generate five ideas for a new high-performance sneaker line.”

→ Output was broad and generic.

Add format: “List concepts and materials in an outline.”

→ Output became structured and usable.

Add context: “For cross-training athletes.”

→ Ideas became relevant to a specific audience.

Add references: Include examples of existing sneakers.

→ Output turned creative and targeted, for example, temperature-regulating uppers inspired by real products.

With each layer, task clarity, format, context, and references, the response moved from vague to actionable.

Shot prompting:

Zero shot: no examples
Single shot: one example
Few shots: two to five examples, often the sweet spot

The more relevant the examples, the more accurate and creative the response.

3) Iteration Methods

Even good prompts miss sometimes. Here is how I sharpen them:

Revisit the framework: Add detail to Task, Context, References, persona, or format.
Break it down: If it feels like spaghetti, split it into steps. Clarity wins.
Rephrase or analogize: New phrasing or a parallel task can unlock better output.
Add constraints: Time, place, style, audience. Narrow the target.

Mini iteration template:

“Refine by adding [persona or audience or format], add [constraints], and tighten the context. Need more precision? Break it down: [Step 1], [Step 2], [Step 3]. Or reframe it as a [story or checklist or case].”

Iteration is not just fixing. It is evolving.

4) Image and Multimodal Prompting

Image prompting basics

Think of image prompting like directing a photoshoot or setting a film scene, only your actors are pixels. The clearer you are, the closer the output will match your vision.

Subject and action: Who or what is in frame, and what are they doing, for example, “a surfer riding a giant wave.”
Style or medium: Realistic photo, cartoon illustration, minimalist vector, choose one.
Composition: Foreground, background, focus, angle, perspective.
Colour and mood: Bright and energetic, soft and muted, dark and moody.
Lighting: Natural daylight, neon glow, and dramatic spotlight; each changes the feel.
Output specs: Aspect ratio and resolution, for example, 16:9 at 4K. Leave negative space if you will add text.
Constraints: Say what to exclude, for example, “no text overlay, no people, minimal background.”

Template:

“You are a [visual designer]. Create an image for [purpose]. Subject or action: [details]. Style or medium: [photo, illustration, or vector]. Composition: [foreground or background or angle]. Colour and mood: [palette and vibe]. Lighting: [type]. Output specs: [aspect ratio and resolution]. Constraints: [no text or people, etc.].”

Multimodal prompting

Combine text, images, audio, or video for richer interactions.

Example: Upload a fridge photo and ask, “What recipes can I make with just these ingredients?” The model interprets the image and suggests options.

Why it works: It feels vivid and practical, closer to chatting with someone who understands context.

The catch: These models still struggle with abstract reasoning and common-sense leaps. Great at interpreting what is there, not perfect at making jumps.

Bottom line: Mixing modes often produces more useful, more human results than text alone.

5) Responsible Prompting

AI can be biased, inaccurate, or inconsistent. It is on us to use it responsibly.

Common risks and how to handle them

Bias and stereotypes: Ask for diverse perspectives, correct assumptions, and request underrepresented voices.
Hallucinations, false info: Fact check and cross-reference with reliable sources. Avoid prompts that invite speculation as fact.
Inconsistencies: If outputs feel random, add examples, clarify phrasing, and tighten constraints.
Copyright and IP: Avoid prompting for trademarked styles or protected content you do not have the right to use.

My workplace AI checklist

✅ Confirm AI is the right tool

✅ Avoid reinforcing bias

✅ Get approval before client-facing use

✅ Protect privacy and sensitive data

✅ Fact check every output

✅ Be transparent about when and how AI is used

Responsible prompting is not paranoia. It is being deliberate.

Key takeaways

Framework: Task → Context → References → Evaluate → Iterate
Iteration, ABI: Always refine. Split tasks, rephrase, and add constraints
Image or multimodal: Same framework, plus style, composition, and specs
Responsible prompting: Humans verify, fact-check, and disclose AI use

👉 Bottom line: Good prompts are rarely one and done. They are built through clear design and steady iteration.

At the end of the day, prompting is a bit like any creative skill: writing, design, or even cooking. You learn the basics, experiment, and mess up a few times, but you get better the more you play with it. I’ve seen firsthand that small tweaks can turn “meh” outputs into something surprisingly sharp.

So don’t be afraid to keep refining, testing, and most of all, having fun with it.