The 6-Step Framework for Pristine AI Prompts 🎯

By the end of this post, you'll know exactly how to prompt your chatbot to get pristine results, using a six-step framework inspired by AI guru Liam Otley. This framework can improve your AI responses by up to 300%, based on validated research, and I've personally been blown away by the results.

So let's get into it.

Most People Are Terrible at Prompting

Most people are terrible at prompting AI. They type in a random question, get a mediocre response, and then complain that the chatbot isn't that good.

Here's the thing: chatbot prompting is such a crucial skill these days that you really need to know how to do it properly. Otherwise, the consequences can actually be quite severe. I learned this the hard way when I was preparing for a job interview and used really bad prompts, got really bad information, and then used that information in the interview. Suffice to say, I didn't get the job.

Don't make the same mistake I did. AI needs work and effort to get good results. Otherwise, it's pretty much a garbage in, garbage out situation.

The 6-Step Framework Breakdown

Each part of this framework serves a specific purpose, and it'll take some time to refine each prompt. But once you've nailed it, you'll get great results every time. Let's go through each component one by one.

1. Role 🎭

This is where you tell the AI exactly who it needs to be. Instead of just asking a question, you're essentially hiring the AI for a specific job.

So rather than saying "write me a marketing email", you'd say something like: "You are a direct response copywriter with 10 years of experience in email marketing."

You could take this even further with something like: "Act as a professional email communication specialist who crafts modern, concise business emails with personality. You understand the balance between professionalism and authentic personal voice, creating emails that feel genuine whilst maintaining business credibility."

I know that sounds like an absolute mouthful, and it might feel insane that you'd need to go into this much detail. But this technique alone can improve performance by 15 to 25%. The AI genuinely performs better when it has a clear identity to embody. This step is a must.

2. Task ✅

This is what most people think of when they interact with a chatbot. They just tell it what to do.

Your task should always start with a verb: generate, write, analyse, create. Be specific about what you want, but don't dump everything here. That comes later.

3. Specifics 💡

This is where things get interesting. There's a technique called emotion prompting, and research shows that adding emotional phrases can improve AI performance by up to 115% on complex tasks.

You might add something like "this is very important to my career" or "I greatly value your thorough analysis." It sounds weird, but it works. Think about it this way: if you gave someone work to do and emphasised just how important it was to you that they got it right, the chances they'd put more effort in are much higher. The same logic applies with AI.

4. Context 🗂️

This is where you explain the bigger picture. Why does this task matter? What's the business situation? What environment is this AI operating in?

The more context you provide, the better the AI can tailor its response to your actual needs.

5. Examples 📋

This uses a technique called few-shot prompting, which can boost performance by around 14%. Basically, you tell the AI exactly what good output looks like by providing three to five examples.

This is incredibly powerful because the AI can pattern match to your preferred style and format.

6. Notes 📝

This is your last chance for fine-tuning everything: formatting requirements, things you definitely don't want, tone adjustments, and any final reminders.

It's like having a conversation with someone and saying "oh, and just one more thing." I'll explain more about why this matters later in the post.

A Real World Example

Let me show you this in action, because seeing the difference is honestly pretty striking.

Here's probably an extreme example of a basic prompt most people would use:

"Write a cold email for my design agency."

And that's it. You'll get a generic, forgettable email that sounds like every other cold email out there.

Now, using our framework, the same request becomes:

Role: Act as an experienced business development specialist for creative agencies with a proven track record of securing high value clients through strategic outreach. You understand the unique challenges small tech startups face and excel at crafting messages that resonate with busy founders and decision makers.

Task: Generate a compelling cold outreach email for a boutique design agency targeting small tech startups, focusing on how exceptional design can accelerate their growth, improve user acquisition, and increase investor appeal.

Specifics: Create an email that feels personalised and research-driven. Keep the tone professional.

Context: This email will be used as a template for outreach to 50+ carefully researched tech startups.

Examples: Here's the tone and structure I'm after. "Hi [Name], I was using [Product] this week and noticed the onboarding drops users right before the aha moment. When we redesigned that exact flow for [Similar Startup], their week-one activation jumped 34%. I've sketched two quick ideas for [Product] - worth a 15-minute call to walk you through them? No pitch, just the mockups." Keep openers specific to their product, lead with a problem they'll recognise, back it with a concrete result, and close with a low-friction ask.

Notes: Research each startup thoroughly. Avoid design jargon and focus on business metrics and outcomes. This email must feel like it comes from someone who genuinely understands their business challenges.

You'll notice some of the same sentiments get rehashed in different parts of the prompt using different wording. That's intentional, and the difference in output quality is genuinely huge.

Limitations to Keep in Mind

This framework works best for complex tasks that require creativity and analysis. For simple questions, the kind you'd ask Google, it's honestly overkill.

Also worth noting: if you're paying per token, longer prompts mean higher costs. So there's a trade-off to be aware of.

A Few More Things That Make This Work

The lost in the middle effect 🎯 There's something called the lost in the middle effect, which can impact performance by up to 56% if you ignore it. AI models pay more attention to information at the beginning and end of your prompt than stuff in the middle. That's exactly why the notes section is so powerful: it's your final opportunity to emphasise what matters most.

Interestingly, this mirrors how humans work too. Scientific research shows people remember things at the beginning and end of a list far better than the middle. So treat your AI not as a machine, but like you would another human or colleague.

Chain of thought prompting 🔗 This is where you ask the AI to think step by step or show its reasoning process, and it can improve results by up to 90%. You can incorporate this into any part of your framework by adding phrases like "think through this step by step" or "explain your reasoning."

Pick the right model for the task 🎛️ A brilliant prompt still needs the right engine behind it. For anything genuinely complex, like deep analysis, strategy, or multi-step reasoning, reach for a more capable, thorough model rather than the fast default. That's exactly where this framework earns its keep, and a lightweight model will often waste the detailed prompt you've carefully built. For quick, simple asks, the faster model is perfectly fine, so save the heavyweight for the prompts that actually warrant it.

Start simple and build up A good approach is to start with a simple version of your prompt, just role, task, and context, then gradually add specifics, examples, and notes to refine the output. Most of my best prompts have evolved over several iterations rather than being written perfectly the first time.

Emotion prompting really does work I mentioned this earlier, and I know it sounds gimmicky, but the research is solid. Phrases like "this is vital to my career" or "I greatly value your thorough analysis" genuinely improve response quality. The AI doesn't have feelings, but it was trained on human text, and those emotional cues seem to trigger more careful, thoughtful responses.

Use markdown formatting Markdown is a simple way to add formatting like bold, italics, lists, and links by typing a few symbols instead of using buttons or menus. It's also how LLM engineers interact with and tune models behind the scenes, so it makes sense to use it in your prompts too. It's an easy way to eke out any final gains.

Wrapping Up

That's the six-part framework that can transform your AI interactions. If you use AI for work, content creation, or problem solving, this structure will genuinely change how effective your prompts are.

Good prompting is a skill. Like any other skill, it gets better with practice. Put the initial work in, and the results won't just be better, you'll save a ton of time on every result going forward.