Writing effective prompts for Copilot

In this chapter

• Understand prompts and responses

• Discover and use suggested prompts with Copilot Lab

• Use the elements of effective prompts

• Write prompts to generate images

• Use more advanced prompting techniques

Practice files

No practice files are necessary to complete the practice tasks in this chapter.

The most important skill to master when working with Copilot is writing effective prompts. Any instruction or request that you give to Copilot is a prompt. Prompts can be simple instructions with a single response, or they can include context, source content, examples, and details of the desired output format. Prompting can also be more than just a single request and response. Understanding how to refine and iterate your prompts will enable you to have a conversation with Copilot that will give you much better results. In this chapter, you will learn how to write prompts to help you work effectively with Copilot with both text and images, using the main elements of prompting. You will understand different prompting techniques and the scenarios in which you can use them. You will also learn how to use Copilot Lab to find suggested prompts to help you get started.

Understand prompts and responses

Prompts and responses are the fundamental components of the interactions you will have with Copilot.

How generative AI works

When you chat with Copilot or ask Copilot to do something for you, it uses a large language model (LLM) to process your prompt and provide a natural language response. A large language model is a type of AI designed to process and generate text. It has been trained on huge datasets to understand how language patterns work and is able to process your request and generate a response by predicting what comes next using these language patterns. Copilot uses the LLM to write text, provide responses, and carry on a conversation in the context of what you’ve requested. Copilot doesn’t actually understand what you are asking, or what it is saying in response, and it doesn’t have any thoughts, opinions, or feelings. It is simply using the patterns it has learned from the training data to predict the most likely words or sentences to generate a response.

Generative AI is creative by design, making different predictions each time you provide a prompt. This means it won’t always generate the same answers, even when you use the same prompt. The more specific you are with your prompt, the more likely you are to get a similar, or even the same response each time. If you ask Copilot to generate content on a narrow or specialized topic, such as “Explain how to fold a paper crane,” you will get almost the same content each time. If you ask Copilot for a well-established fact that doesn’t change, such as “What is the capital of France?” you will also get a consistent answer. If you work with more general or creative prompting, such as “Write an outline for a detective story,” the generated responses will be very different each time.

Generative AI has a “temperature” control that determines the scale of how predictable or creative the response will be. Lower temperatures mean the response will closely follow predictable, learned patterns, whereas higher temperatures encourage more creativity and innovation in the response. When you use Copilot on the web and mobile experiences, you can choose whether you want Copilot to respond in a way that is more precise, more balanced, or more creative. Using the more precise option will give you more factual and consistent responses, whereas the more creative option can be more likely to fabricate information, known as hallucinations. The more creative option, however, will give you better results when you want to use language in a creative way, such as writing a poem or thinking about a question or problem in different ways.

Generative AI is by nature creative, even at a low temperature or using the more precise option. It is therefore important to expect that as you use Copilot to generate drafts or rewrite content, that can yield a different response even if you use the same prompt. If you get an answer you like, make sure to accept or save it, or copy and paste it into a document or notes for future use. This creative aspect also means that if you don’t like the response to your prompt, one thing you can try is submitting the same prompt again to regenerate a different version of the response. In most cases, however, learning how to write better prompts and refine your instructions through a conversation with Copilot will be the way to get the best results.

This chapter teaches you the skills you need to write effective prompts in a range of different scenarios with Copilot.

The best use cases for Copilot

One of the biggest challenges with adopting Copilot is learning to think differently and understanding how and where generative AI can provide value to enhance your creativity and productivity. When you first start using Copilot, you are likely to get some disappointing or unexpected results. It is important not to give up. Think about Copilot as your virtual assistant and consider when and how to use Copilot in the same way you would if you were delegating a task to a junior assistant.

The easiest way to get started is to think about how Copilot can free up your time. Think about tasks that take up a lot of your time, or tasks that are tedious, which don’t need the full power of your knowledge or creativity to complete. For instance, you could ask Copilot to help you put together a meeting agenda based on your notes.

Copilot can also help you rewrite or repurpose your work, where you have already done the thinking or creative work. This can include turning a document into a presentation, rewriting it for a different audience, or writing an executive summary.

You can go beyond productivity tasks with Copilot, using it to help you solve problems or brainstorm new ideas. You can also reverse the pattern of the conversation, asking Copilot to ask you the questions to help you think differently, or to challenge your assumptions on a topic.

Prompting in the web experience vs. app experience

With Copilot Pro, you can chat with Copilot on the web and in mobile apps, as well as in Microsoft 365 applications. These experiences provide you with different ways to work with Copilot.

When working with Copilot in the web or mobile experiences, you can have a more free-form conversation. This is a great place to start with idea generation, problem solving, or brainstorming. This is also the best place to work if you want to have a conversation with Copilot in which you go back and forth and refine your ideas and requests, or to use more advanced prompting techniques covered later in this chapter, such as the persona pattern. Copilot on the web or mobile is also the best place to generate images.

Think about Copilot in the Microsoft 365 applications as a series of “specialist” assistants that are designed to help you with the kinds of work you do in each application. For instance, Copilot in Word is very good at helping you draft and rewrite documents, whereas Copilot in OneNote is great at identifying action points and creating to-do lists and summaries from your unstructured notes.

When you work with Copilot in the Microsoft 365 applications, the prompts you write will mostly fall into one of these task areas:

• Create prompts ask Copilot to create a first draft of something for you, like a document, paragraph, meeting agenda, presentation, column, or email.

• Edit prompts ask Copilot to rewrite your content in different ways, changing the length, tone, or purpose, including giving you variations to choose from. Edit prompts can also help with organizing your presentations and applying conditional formatting to your spreadsheets.

• Ask prompts help you find specific information, coach you to create better content, or answer questions about your content or related topics.

• Understand prompts help you summarize your content, find insights, and extract to-do lists or key points.

You can get the most out of Copilot Pro by using the web and application experiences together. For instance, you can start by brainstorming and working through a new idea in Copilot on the web, and then use the output of that as a prompt to draft a new document in Word. You can even ask Copilot on the web to help you write or improve your prompts.