layer 1 layer 2 layer 3 layer 4 layer 5 abstract shapes

The power of prompt engineering: how to get the best Copilot results

Posted 05th June 2024

What do you think a CPO does in a business?

Are they the Chief Product Officer? Or does that middle initial stand for Planning, or even Process?

Well, now there’s a new P to add to the mix: the Chief Prompt Officer. While it isn’t a role you’ll see dominating every vacancy list just yet, it’s a title we’re hearing more often as organisations seek to extract maximum value from generative AI assistants such as Microsoft 365 Copilot.

In this article, we’ll define – and consider the growing power of – ‘prompt engineering’ for business. We’ll also look at ways that your teams can create better prompts and overcome using bad ones; and show how Copilot itself can help to make your prompts better.

The power of better prompts in Microsoft Copilot

Put simply, prompt engineering makes Copilot more effective and more efficient. A well-crafted prompt – the question or command a user inputs into a generative AI application – provides the link between human and machine. Stating clear requirements that a human would be able to process is always a good starting point.

To get slightly more technical, prompt engineering gives the developers working on AI applications more control and scope to tailor responses. These will be more likely to return exactly the information a user is trying to discover.

In other words, thanks to prompt engineering Copilot can understand user intent and deliver relevant results sooner, without the need to repeat or refine requests, or waste time doing so. The quality of the prompt dictates the quality of the results.

Of course, the flip side of this is a ‘bad’ prompt. Often, that means the user’s input is too general and they will end up wading through copious content to locate what they need – if they can find it at all.

So, if prompts can be a valuable way for your team to save time and enhance their work, by incorporating correct information they wouldn’t previously have known, what do users need to do to ensure they’re getting the most out of Copilot?

How to create effective prompts in Copilot

As mentioned, it’s generally the case that a well-written prompt, drafted with clear and specific parameters, leads to more useful responses.

Here are a few top tips on creating more effective prompts. Users should try to include:

  • Context – The more context you can add, the better e.g. why do you need the information and how do you plan to use it? This goes further than entering a simple search term into Google
  • Expectation – Should the response ideally be tailored for a certain target audience, or in a specific format?
  • Source – Can the user add a particular place Copilot should search to locate the required information?
  • Address – Tone matters! Calling Copilot ‘You’ in a prompt is more effective than writing ‘the AI’ or ‘the assistant’

These are some of the basic steps users can take to refine the way they interact with AI and achieve better outcomes. But there are other aspects to be aware of that can make a big difference to the quality of prompt results.

Many of these involve avoiding ‘hallucinations’ by the AI. This is when the tool seems to supply made-up information that has no basis in reality but is presented as fact. It can happen for a number of reasons, which include:

  • Prompts that are too ‘loose’, in other words not specific enough to construct boundaries for the AI
  • Use of local dialect or idioms in the prompt might mean AI has to guess your meaning, and is more likely to present responses that don’t make sense
  • Prompts that are written to deliberately mislead or ‘attack’ – be overly negative towards – the platform

Taking Copilot’s advice on prompts

Everyone should be able to write a great prompt. The good news for Copilot users is that the tool can help draft and revise prompts till they’re the best they can be.

Copilot can rewrite a prompt to provide more actionable results. Simply clicking on the ‘edit’ icon after entering the text of an initial prompt will refine it within seconds.

For example, if a user enters “Create FAQ document for [subject x]”, the edit function will lengthen the prompt but also make it more specific and useful: “Create an easy-to-read FAQ on [subject X] including clear headings and bullet points, with simple questions followed by comprehensive answers.”

Here’s an example of how a great prompt could look:

Copilot, I’d like your help to wrap up my day, and prepare for tomorrow.

Please name the first section: ‘Today.’ Can you please write a short paragraph of my work today, and separately list tasks that came from my meetings? Also highlight my important actions from my email based on your analysis.

Please name the next section: ‘Your Meetings Tomorrow.’

Then, can you please help me with an overview of my meetings tomorrow. Create a table, and list who organised the meeting. Add details of the meeting. Please also show the category you would place that meeting in (personal, internal, external or traveling) and add any important things I need to know about the meeting.

Please show me what % of time I am spending on the different categories in a list.

Please name the last section: ‘Get prepared.’

Finally, as my personal assistant, please write a small summary and brief me on all of the important things I need to know to make tomorrow a productive day. Also, please suggest when I should consider taking breaks.”

Start a ‘prompt library’ for better results

Meanwhile, Copilot Lab allows users to refer to and incorporate suggested prompts based on those which their colleagues have already used successfully.

Prompt Buddy is one way to provide this opportunity. A free Microsoft Teams app, it’s a collaborative space where team members can share their favourite AI prompts and even upvote the ones they think are best. This library of prompts provides a perfect platform for the CPO – or role responsible for prompts – and can be tailored to individual organisations. It’s preloaded with Copilot categories but includes the functionality to add new ones.

Prompts have the power to unlock value in your users’ daily work and therefore your organisation as a whole.

It’s worth thinking about designating one or more team members as your organisation’s ‘prompt champion’, overseeing and advising on prompts across the business. In this regard, Copilot can be a valuable companion for AI adoption and training; not least as you pilot the use of prompts across your organisation.

If you need further support to implement Copilot effectively in your business and help your users get better results from their prompts, get in touch