Mastering the Art of AI Prompt Engineering

How to effectively use AI prompt engineering for your organisation.

Large Language Models (LLMs) sit within a branch of artificial intelligence called Natural Language Processing (or NLP). They are deep learning models that have been trained on large text datasets from books, websites, social media, chat logs, and multiple other sources, so they can learn to predict the next word in a sequence and generate coherent and human-like text. 

Of course, many of us will know and use LLMs already – the likes of Chat GPT, Codex (used by Microsoft Copilot), and Google’s BERT, for example. Indeed, in this new era of artificial intelligence, where we must learn to collaborate and strategise alongside AI, rather than instead of it (and, indeed, rather than simply deferring to it), being able to effectively AI prompt engineer, and with the right guardrails in place, really is the key to unlocking its potential. 

2025 is going to be a big year—especially for how we communicate with AI. If text was the first modality, and voice the breakthrough of 2024, I think vision will be the standout for the year ahead.

Mustafa Suleyman, CEO of Microsoft AI 

Balancing human and machine creativity

Whether you’re writing an article, developing code, or searching for new solutions to business challenges, knowing how to craft effective AI prompts can make all the difference.

Having a clear, specific goal in mind is crucial, as a clear understanding of the context of your prompt, and how to tailor it to guide the AI towards the desired outcome. By providing relevant details, setting clear parameters, and avoiding ambiguity, you can ensure that the AI generates responses that are both accurate and actionable.

You can put as little or as much information as you wish into your prompt, but the most effective prompts are likely to include the following:  

A clearly defined objective: begin with a clear task or goal in mind 

Example: “Generate ideas for automating repetitive tasks in IT infrastructure management.

Context: include relevant details to guide the AI 

Example: “Generate ideas for automating repetitive tasks in IT infrastructure management, focusing on tasks such as server monitoring, patch management, user provisioning, and backup scheduling. Include examples of tools that can be used to implement these automations.”

Specific language: frame the prompt with clear, concise wording to remove ambiguities 

Example: “Generate a detailed list of automation ideas for IT infrastructure management, specifically for tasks including server uptime monitoring, and automated security patching for Windows servers, onboarding new users with predefined access levels, and scheduling incremental database backups. For each task, suggest a tool or framework, provide a brief implementation outline, and highlight any potential challenges or considerations.”

Parameters (optional): specify constraints like tone, format, length, or direct AI to source information from specific places 

Example: Using the above prompt we could specify and add-in parameters such as: “Limit suggestions to widely used tools such as XX and XX.”  Or “Provide an implementation outline in no more than 150 words per task.” Or “Highlight up to 2 key challenges or considerations for each automation idea.”

Remember, iterative refinement is a crucial part of working with AI, as it helps to fine-tune responses, making them more relevant to the task at hand. Adding or modifying parameters, clarifying ambiguous instructions, or providing more examples can guide the AI toward a better understanding of what is required. By continually testing and adjusting your prompts, you improve not only the AI’s output but also your ability to communicate more effectively with it, resulting in more accurate, actionable, and tailored responses.

Experimenting with prompt styles

There are different styles of prompts, each suited to specific tasks:

• Open-ended prompts: Useful for brainstorming and creative exploration. Example: “What are some innovative uses for virtual reality in education?”
• Directive Prompts: Great for obtaining specific outputs. Example: “Create a detailed market analysis report for a mid-size e-commerce company specialising in sustainable fashion.”
• Comparative prompts: Useful for evaluations or comparisons. Example: “Compare the advantages and disadvantages of adopting a subscription-based pricing model versus a one-time purchase model for [specific software]”.

Pushing this a little further, users may also wish to experiment with role-based prompts, where one asks the AI to assume a role to retrieve a more focused response. For example, “You are a professional editor. Review the following text for grammar and style improvements.”

Remember, for complex tasks, it might make sense to break prompts into smaller steps over multiple interactions, expanding on certain aspects of the response as you see fit. For example: “Give me an outline for an essay on climate change.” And then: “Expand on the second point in the outline.”

Collaborating with AI: shaping the future of creative and professional work

As Mustafa Suleyman stated in this article’s opening quotation, treating AI as a collaborator, rather than a mere tool, represents a shift in how we interact with technology.

However, instead of viewing AI-generated responses as final answers, it’s best to consider them as foundations to build upon — springboards for creativity, innovation, and problem-solving. This is because AI-generated responses are inherently based on patterns, probabilities, and existing knowledge, which makes them powerful tools but not infallible solutions.

Whilst this dynamic partnership has the potential to transform industries and redefine how we approach tasks in the future, it’s important we remember the limits of AI, including obstacles such as:

• Limited context awareness: AI models lack true understanding and often rely on the input provided. They can miss subtle nuances or unique factors specific to a problem, especially in complex or highly contextual scenarios.
• Potential for error or bias: AI systems are trained on large datasets, which may contain outdated, incomplete, or biased information. This means their responses could inadvertently reflect inaccuracies or skewed perspectives.
• Lack of human judgment: AI lacks intuition, ethics, and emotional intelligence. Human judgment is crucial for interpreting, validating, and contextualising AI suggestions to ensure they align with broader goals, values, and strategies.
• Fostering creativity: AI excels at generating ideas or summarising information, but human creativity is essential for refining, adapting, and innovating beyond the patterns AI identifies.
• Complexity and ambiguity: Some problems require deeper critical thinking, cross-disciplinary insights, or ethical considerations that AI cannot fully address. Humans are better equipped to handle ambiguous or open-ended challenges.

As we embrace AI collaboration, then, it still makes sense for focus to remain on human oversight and values. While AI can generate content, humans provide critical judgment, context, and emotional intelligence — qualities that AI cannot fully replicate. Balancing the strengths of AI with human creativity ensures that the future of collaboration remains ethical, inclusive, and innovative.

Download our AI prompt engineering cheat sheet here.

Final word

Remember, effective AI prompt engineering is both an art and a science. It requires clarity, specificity, and a bit of creativity. By following the above tips and techniques, you can harness the full potential of AI tools, turning your ideas into reality with ease.

I encourage users to experiment, learn, and refine — the possibilities are limitless once we master the art of prompting. Whether you’re looking to boost efficiency, generate creative ideas, or solve technical challenges, a well-crafted prompt might well be your first step toward success.

If you’d like to discuss AI prompt engineering or how AI in general could transform efficiency for your organisation, or learn more about implementing Microsoft 365 Copilot or Copilot for Security, please get in touch using the button on this page.