Prompt engineering 101: How to write an AI prompt that actually works  

Introduction 

Ever asked an AI tool for something, only to feel like it’s not quite understood the point?  

Yep, we’ve been there too.   

Generative AI is already transforming our professional and personal lives, but it’s only as clever as the prompts you feed it—and getting those prompts right can sometimes feel like guesswork. 

We’ve touched on prompt engineering before here on the Get Support blog, but today we’re going a bit deeper with a detailed guide on how to write prompts that actually work.  

What is prompt engineering? 

Prompt engineering is all about carefully crafting the instructions (a.k.a the prompts) that you feed into AI tools like Microsoft 365 Copilot or ChatGPT to get truly useful outputs.  

But there’s more to it than just typing in a few words and hoping for the best. Good prompt engineering means being thoughtful and precise about exactly how you word those prompts.  

Consider prompt writing a bit like giving instructions to a colleague who’s totally new to your business. If your instructions are too vague, they’ll probably produce generic or irrelevant results. But if you clearly outline the task, the context, and exactly what you’re expecting, they’ll provide exactly what you need (or at least get much closer!). 

At a practical level, good prompt engineering involves: 

1. Clearly defining your goal: Know exactly what you want from the AI before you ask. 
2. Establishing a role: Direct the AI by specifying the persona or perspective it should adopt (e.g., IT expert, marketing specialist, HR advisor). 
3. Providing the right context: Offer relevant background information so the AI gets the full picture. 
4. Specifying the output format: Guide the AI to respond in bullet points, numbered lists, short paragraphs, or even conversational tones, depending on your needs. 

The clearer your prompts, the better (and more practical) the responses you’ll get from your AI tools. 

Why some prompts just… aren’t very good 

You’ve sent a prompt to an AI tool, but the result isn’t what you expected, so you ask for an adjustment, but the next response is even further away from your goal.  

Rinse and repeat. 

It’s easy to assume that the issue here lies with the AI itself, but that’s not always the case.  

Many prompts just aren’t specific enough or leave far too much room for interpretation. When you ask something broad like, “Give me cybersecurity advice,” you might as well be asking a stranger for directions without telling them where you’re trying to go. 

Another common issue is the lack of context. AI can’t guess your company’s industry, your particular business challenges, or the audience you’re trying to reach. Without you defining that context, the AI is left to make assumptions that might not align with your actual needs. 

So how do we put this right? 

How to write prompts that actually work 

Crafting an effective AI prompt doesn’t have to be complicated, but it does require some thought and preparation.  

Here are five essential tips to keep in mind next time you’re trying to craft that perfect prompt:  

1. Give your AI a clear role to assume. Treat your AI like an actor you’re directing by clearly stating the persona it should adopt. So, instead of “Write an email,” say, “Act as an HR professional drafting a friendly staff email about our updated leave policy.” 
2. Always be specific. Vague prompts give vague responses. Rather than “Give me cybersecurity tips,” ask specifically: “Provide three actionable cybersecurity tips for small businesses using Microsoft 365.” 
3.  Include meaningful context. Context improves relevance dramatically. Rather than a generic question, add specifics like: “My small UK business with remote staff recently had an email breach. Suggest three immediate cybersecurity measures for Microsoft 365.” 
4. Break complex questions down. Complex questions are best handled in bite-sized pieces. Instead of one long, complicated question, split it logically: “Suggest three Excel functions for managing a monthly budget.”, then, “Recommend an Excel formula to calculate monthly profit margins.” 
5. Specify your preferred format. Clarify exactly how you’d like information presented in the AI’s response. Clearly request your preferred style, such as: “List three cybersecurity best practices as bullet points.” This makes sure outputs as close to ‘ready-to-use’ as possible even after one generation.  

Examples of good prompts in action 

We know theory isn’t always the best way to learn, so let’s put some of this into practice. 

Here are some real-world examples of some not-so-good prompts and some much-much-better prompts. They should help illustrate how to put the pointers above into action: 

Poor prompt ❌	Improved prompt ✅
“Help me write an email.”	“Act as a customer support specialist at an e-commerce store specialising in eco-friendly clothing. Draft an empathetic email to a customer requesting a refund 10 days past our 30-day return window.”
“What are good Excel tips?”	” Assume you’re a financial analyst with 10+ years’ experience in UK small business consulting. Provide three essential Excel formulas to track and forecast monthly budgets.”
“Explain cybersecurity.”	“You are an IT consultant specialising in Microsoft 365 for small UK businesses under 50 employees. Recommend three immediate, budget-friendly actions to improve cybersecurity.”
“Give me marketing ideas.”	“Act as a digital marketing expert. Suggest three creative yet affordable marketing ideas suitable for small UK businesses selling products online and targeting new parents.”

Ready to do more with generative AI? 

It might sound complex, but prompt engineering is really just about clear communication.  

And the great thing about generative AI is that it uses natural language, so there’s no coding to worry about. Talk to your AI tools like you’d talk to a person and you’re likely to see much better results—that’s the way they were designed.  

If you’d like to learn more about Microsoft 365 Copilot or how to bring AI into your everyday workflows, just ask your Get Support Customer Success Manager now or call the team on 01865 594000.