Copilot 101: The importance of prompts and how to create great ones

Good communication is paramount to success. Of course, we all know this applies when we’re speaking to our colleagues, clients, and suppliers. But what about when it comes to your AI assistant?

With the advent of advanced AI software like Microsoft Copilot, the ability to articulate requests clearly is becoming ever more crucial. Well-crafted prompts can significantly impact the accuracy of responses generated by AI assistants, streamlining processes and saving you valuable time.

In this article, we’ll explore what makes an effective prompt for Microsoft Copilot, provide real-world examples that you can replicate, and share practical tips to help you improve your prompt-writing skills.

What is a prompt?

A prompt in the context of AI is the same as it is in real life – a query or instruction given to obtain a specific response or action. Much like you might give a colleague or friend a prompt to do something for you or help you with a task, you can do the same for your AI assistant. Of course, your AI assistant probably can’t make you a cup of tea (at least not yet), but you can give it business or productivity prompts that you might give your assistant or a team member.

What makes an effective prompt?

We’ll be using Microsoft Copilot as our AI assistant, but these general rules for effective prompts should also work with other AI programs, such as ChatGPT.

For a prompt to work, you need to give Copilot a goal – define the objective of the prompt. This is the bare minimum, and Copilot will usually return you with an answer, albeit of varying value. Some examples of this might be: “give me ideas for meals”, “help me plan a party”, or “write me a meeting plan”. As you can probably tell, these are very vague, generic prompts, but it can be helpful to see what Copilot returns from these prompts so we know where to improve.

I asked Copilot to “write me a meeting plan” with no extra context, and this is how it responded:

Given how vague the prompt was, Copilot's response is quite impressive. Copilot has scanned my Microsoft environment, including my emails, Teams chats, and files, in an attempt to provide relevant information in answer to my prompt. This includes things like likely attendees, past meetings I can review for inspiration, venues, reference materials, and other 365-integrated programs that might help with preparation.

Whilst a good attempt from Copilot to make something of our poor prompt, it’s unlikely that this response is what you’re looking for.

Copilot has helped us map out how we might run a meeting, but it hasn’t given us anything in terms of the actual meeting plan itself. A vague prompt returns a vague answer, and so we should provide more detail and be specific about what we want. We need four key elements to create a strong, specific prompt to generate the answer we’re looking for:

1. Goal
2. Source
3. Context
4. Expectation

Goal

We’ve already got a rough goal, but it can always be improved. Think about the specific result you want from Copilot – “a meeting plan” is fine, but it’s open-ended, and there’s no way of telling what response we’ll get. For example, for a 30 minute meeting slot, you don’t want 15 points of discussion.

We’ll change it to the following to further refine it, but still give Copilot some wiggle room if it wants to expand:

"Write me a meeting plan consisting of 3-5 discussion points."

Source

Secondly, we want to consider our source. Providing a source tells Copilot where you want it to draw information from. As mentioned, Copilot is unique because it integrates with your Microsoft 365 environment, so it can draw information from specific programs if you wish. It can also search the web for data and information.

Let’s say our meeting is the upcoming marketing meeting on a new campaign – we’ve talked about it before over Teams and email, but now we want to get a proper plan together. The prompt might now look something like this:

"Write me a meeting plan consisting of 3-5 discussion points. Use Teams chats and emails between the marketing team for information, and also search the web for anything that might be useful.”

Context

Thirdly, we want to add some context. What’s the purpose of the meeting? As mentioned above, it’s for a new campaign and we want to put together a proper plan.

Give Copilot that information to ensure it provides a relevant answer:

“Write me a meeting plan consisting of 3-5 discussion points. We are planning a new marketing campaign for the rollout of Microsoft Copilot to our clients over the next 6 months, and the next marketing meeting will be used to discuss our plan. Use Teams chats and emails between the marketing team for information, and also search the web for anything that might be useful.”

Expectations

It's already looking pretty good, but we can add one final touch. By stating our expectations, Copilot can respond in the best way possible to fulfil our request. This includes anything from the type of language used to how much detail you want for each discussion point. We’ll include these two ideas in the prompt before we send it to Copilot:

“Write me a meeting plan consisting of 3-5 discussion points. We are planning a new marketing campaign for the rollout of Microsoft Copilot to our clients over the next 6 months, and the next marketing meeting will be used to discuss our plan. Use Teams chats and emails between the marketing team for information, and also search the web for anything that might be useful. Please use simple language so everyone can understand, but use marketing and sales terminology where appropriate. Please provide a small amount of detail for each discussion point.”

The new prompt

Whilst this prompt is long, it provides sufficient detail for Copilot to provide us with a relevant starting point. The prompt returned this answer:

As you can see, Copilot provided a detailed response that fit the criteria in our prompt. There are 3-5 discussion points related to the campaign, drawn from the sources we specified and even linked to specific documents.

Whilst this response from Copilot is great and an excellent springboard for discussion, it can always be improved. It is unlikely that you will get the perfect response from just one prompt or message, but taking the time and effort to craft a great initial prompt helps set the tone, topic, and expectations for the rest of the conversation.

I think the response Copilot has given will help the marketing team to form a campaign plan in the next meeting, but maybe I want more info about the content or some examples on certain points – we can ask Copilot this too. It’s important to remember that Copilot chat is iterative – you’re expected to have a conversation with it to refine the responses further.

Whilst the suggested method of goal, source, context, and expectations will always help you create a strong first prompt, the best way to improve your prompting and the responses you get from Copilot is simply to experiment. Have several conversations with Copilot, embrace the back-and-forth, speak to it like you would to a colleague or executive assistant in your normal language, and learn how you can make it work for you!

Tips and Tricks for Copilot Chat

• Keep it conversational. Like humans, AI won’t get it right the first time every time. Have conversations with Copilot – send multiple prompts to improve the response, give it constructive feedback, and mould the program to adapt to your preferences.
• Use positive language. Copilot and other AI tools respond better to positive language than negative language. Tell Copilot what you want it to do, rather than what not to do.
• Review and verify responses. Occasionally, Copilot responses can include incorrect content, due to the vast and diverse nature of LLMs. Evaluate Copilot’s responses and cross-reference with trusted sources when needed.
• Create your own prompt library. Each time you create a new prompt that worked especially well, consider saving it in a document for later use. That way you can reuse the prompts or tweak them for future conversations, share effective prompts with your colleagues, and review and improve them in the future.
• Experiment with Copilot. Keep trying new things and see what works and what doesn’t. For example, you might try asking Copilot for examples from different viewpoints, or maybe try telling Copilot what job role it has for a particular conversation (e.g. “You are a marketing executive”).
• Don’t change topics abruptly. Whilst it’s tempting to use Copilot chat as a catchall for your stream of consciousness, it’s better to start a new chat if you want to discuss a new topic. Abrupt topic changes can confuse Copilot, as each conversation tends to springboard off the initial prompt – if you’ve told Copilot it’s a Marketing Executive and then start asking HR questions, you may get some interesting answers!
• Use Copilot in a respectful, ethical, and lawful manner. Avoid using Copilot for any purpose that might cause harm to yourself or others. See Microsoft’s responsible AI principles and standards.

Helpful resources for Copilot prompts

• Microsoft’s Copilot Lab page – start here!
• Learn about Copilot prompts
• How to write ChatGPT prompts: your 2024 guide - similar principles apply for both ChatGPT and Copilot
• How to write effective prompts for Copilot: Best Prompting Practices
• Found any good resources yourself? Share them with us!