Review prompting best practices

10 minutes

Microsoft built Copilot for Microsoft 365 upon Large Language Models (LLMs) that are connected to your Microsoft 365 apps and data. With Copilot for Microsoft 365, you can go beyond what you can do with other LLM-powered chatbots. You can also get data from Microsoft 365 Apps and your internal data, such as articles, reports, emails, presentations, and more. With Copilot, you can create or edit content, ask questions, summarize information, and catch up on things.

Copilot for Microsoft 365's performance often depends on the clarity and specificity of the prompts provided by the user. Copilot prompts are instructions or questions you use to tell Copilot what you want. As a general rule of thumb, you should expect some back-and-forth conversation to get the results you’re looking for. Remember that interacting with Copilot for Microsoft 365 is an involved process. If the initial response isn't satisfactory, you should consider applying the following suggestions:

Refine the prompt for clarity.
Be more specific or adjust the phrasing.
Use the feedback from previous responses to guide your next prompt.

The following sections provide best practices for formulating prompts, allowing you to get the most out of Copilot's capabilities.

Provide clear and concise prompts

When interacting with Copilot for Microsoft 365, it's crucial to be both clear and concise. While Copilot is designed to understand context, reducing ambiguity in your prompts can lead to more accurate results.

Clear. Make your request or query unambiguous.
Concise. Keep it brief without sacrificing clarity.

Providing a clear and concise prompt is fundamental to harnessing the true potential of Copilot for Microsoft 365. The more specific and detailed your instruction, the better Copilot can align its response to your intended outcome. As illustrated in the following examples, moving from a generic to a comprehensive prompt can drastically improve the quality and relevance of Copilot's output, ensuring you receive content that perfectly matches your requirements.

Initial prompt	Potential issues	Revised prompt
Summarize this meeting transcript	May not capture all relevant details or may omit critical information	Analyze this meeting transcript and summarize the meeting, especially all the key points made by Susan.
Rewrite this document to be more concise	Lack of context on which sections to prioritize or maintain	Rewrite the introduction and conclusion of this document to be more concise while preserving the main points.
Write a personal narrative about a memorable experience	Could generate a narrative that lacks depth or detail, possibly producing a generic story without emotional resonance or vivid imagery.	Write a personal narrative about a memorable experience you had in nature, using vivid sensory details and emotions.
Summarize the product reviews from our latest launch	Could result in a too-general summary that doesn't provide actionable insights	Summarize the top three positive and top three negative points from the product reviews of our latest smartphone launch.
Draft a response to this customer complaint	Too generic; the tone, solutions offered, or the depth of the response may not match company standards or the complaint's severity	Draft a polite response to this customer's complaint about shipping delays, offering a 10% discount on their next order.

Besides being clear and concise, another best practice that can help improve response quality is to provide positive instructions. Microsoft built Copilot to take action, so telling Copilot what “to do” rather than what “not to do” is more effective. Try using “if-then” instructions.

Finally, remember that AI excels at imitation. Large Language Models work by mimicking human conversations, so try to give examples of what you’d like the output to look like. For example: Write a catchy slogan for a new brand of toothpaste that whitens teeth and freshens breath, using the following example for inspiration: “Good things never change.”

Experiment with different styles

Get creative and experiment with different styles to home in on answers that fit your needs. Try specifying tone (neutral, casual, professional) or giving guidance for what kind of language to use. For example, "Use language a nontechnical person could understand."

Analogies, poems, and even historical allegories - such as "What is a moment in history I can use to explain the central message in this doc?" - can also be useful ways to help you process the information.

Give Copilot a point of view from which to answer

When you provide a point of view from which to answer, it helps Copilot understand the context of your request and generate more accurate responses. Usually, that involves some explanation about who you are and what you’re trying to achieve so the AI can roleplay. For example:

You're a social media manager writing LinkedIn copy.
You're a product marketer working on a new campaign.
You're a coding tutor who is great at explaining Python to students.

You can also ask for a response in the style of a specific persona or approach, such as:

Tell me how to solve this problem with the expertise of a Stanford business professor.
Teach me about this esoteric company concept in a way a nontechnical person could understand.

Know what to avoid when creating a prompt

Certain practices can lead to less optimal interactions with Copilot for Microsoft 365. Here's what to be cautious about:

Vagueness. Avoid overly general prompts, which can lead to broad and less useful answers.
Over-complication. Don’t cram too many questions or requests into one prompt.
Assumptions. Don't assume Copilot has the context from previous interactions unless within the same session.

Understand Copilot's limitations

It's essential to recognize that while Copilot for Microsoft 365 is powerful, it's not infallible. Knowing its limitations can help you create effective prompts. For example:

Break down complex tasks. If a task requires intricate thought processes or numerous steps when done manually, it's likely to be just as challenging for Copilot. Consider segmenting these tasks or thoughts into more manageable pieces.
Optimize for repetition. Copilot is more efficient at performing simple repetitive tasks multiple times compared to manually performing the same task.
Ambiguity. Copilot can sometimes misinterpret ambiguous prompts, so always aim for clarity.
Randon responses. Responses you get from Copilot are random. Using the exact same prompt multiple times can result in different responses.
Unacceptable responses. Large Language Models provide the foundation upon which Copilot is built. LLMs can occasionally generate biased, offensive, harmful, or incorrect content. Therefore, it's imperative that you consistently review and validate the responses for accuracy and appropriateness.

Be polite

Using basic etiquette when you interact with generative AI helps generate respectful, collaborative outputs. Using polite language sets a tone for the response. Generative AI also mirrors the levels of professionalism, clarity, and detail in the inputs you provide.

When you work with Generative AI, consider the well-known expression commonly referred to as the Golden Rule. While the wording of the Golden Rule can vary across different cultures, religions, and philosophical traditions, its basic premise is: “Treat others the way you want to be treated.” But how can this advice apply to AI? It’s not as if your Copilot feels appreciative when you say "please" and "thank you." Maybe so, but using basic etiquette when interacting with Copilot helps generate respectful, collaborative outputs.

Large Language Models - also known as Generative AI - are trained on human conversations. In the same way that your email autocomplete suggests a likely next word or phrase, LLMs pick a sentence or paragraph it thinks you might want based on your input. Put another way, it’s a giant prediction machine making highly probabilistic guesses at what would plausibly come next. So when Copilot recognizes politeness in a request, it’s more likely to be polite back. The same is true of your colleagues, strangers on the street, and the barista making your frothy soy latte with a robust espresso, sugar, and garnished with ground cinnamon: When you’re kind to them, they tend to be kind to you too. The same is true with Generative AI tools such as Copilot. It mirrors the levels of professionalism, clarity, and detail in the prompts you provide.

Rather than order Copilot around, start your prompts with “please,” such as:

Please rewrite this statement more concisely.
Please suggest 10 ways to rebrand this product.

Say "thank you" when it responds and be sure to tell it you appreciate the help. Doing so not only ensures you get the same graciousness in return, but it also improves the AI’s responsiveness and performance.

Be curious

While curiosity may have killed the cat, asking preliminary questions is important when working with Copilot. It helps you to provide specific instructions to Copilot and ask the right questions, which in turn helps you to get the most out of Copilot and avoid common pitfalls. Being curious not only helps you to understand what you don’t know, but it also helps you have a growth mindset, which is essential when working with AI.

When you're curious, you're more likely to ask questions that help you to understand the problem better and provide specific instructions to Copilot. By doing so, you help Copilot generate more relevant and creative responses.

Iterate, iterate, iterate

One of the most common mistakes people make with AI is giving up after a single prompt results in a less-than-ideal response. Working with Copilot is a dynamic and interactive process, where you and your Copilot work together to create content, solve complex problems, and learn new information based on your goals, expectations, and feedback. And your initial prompt is just the beginning.

Your initial ask is just a starting point. As you provide more details and context, that conversation helps Copilot understand what you want and generate more relevant outputs. It also helps everyone learn. When you ask Copilot to explain, summarize, or paraphrase something, you can gain new insights and perspectives.

For example, give Copilot feedback when it produces good or bad outputs. Doing so enables it to learn from your expectations and improve its performance. For example:

This sentence is too vague. Please be more specific.
Thank you. This version is better.
Please add a sentence that summarizes the main point of the paragraph.

Let’s say you need to prepare for a presentation. You might first type: Please create a storyboard for a 60-minute keynote where I need to get a room full of salespeople engaged about our new approach to sustainability. Use these files as background.

Once Copilot shares a draft, you can follow up. For example:

Add a key message to land for each section.
That plan is great, but I’d like more detail on why a salesperson would care about this issue.

Keep going until you’re satisfied. Copilot doesn’t get discouraged, so ask for plenty of options. For example:

Give me 10 ideas for an opening paragraph.
Rewrite this section in five distinct styles.
Give me three analogies that might resonate with this audience.

Experimentation is key. Get creative with different prompts, feedback, and refinements until you find what works best for what you need.

Exercise: Compare prompt results using Copilot in Word

The following tips offer more guidance to help you create effective prompts. Each tip includes an exercise that you can perform in Microsoft Word. Each exercise compares the results from different prompts that are associated with the respective tip.

Tip: Include details. In addition to your goal (what you want Copilot to do), consider including some context and explain how Copilot should respond to fulfill your request. Also include any specific information source Copilot should use (such as a file or email messages).

Exercise: Try the following two prompts using Copilot in Word and compare the responses you get. Which prompt gave you a richer and more concise blog post draft?
- Prompt 1: Write a blog post about sustainable practices in agriculture.
- Prompt 2: Craft a 1500-word blog post for a general audience interested in sustainability, focusing on the significance and benefits of sustainable agricultural practices like organic farming and agroforestry. Include real-world examples, innovative technologies, and insights from reputable sources. Conclude with a reflection on the importance of these practices and a call for collective efforts to embrace them
Tip: Structure your prompts for better results. When it comes to prompts, order matters. The order of your instructions can affect the response you get. Later parts of a prompt are likely to be emphasized more than earlier parts. Experiment with different orders to understand the effect. If you want Copilot to use specific files or sources, put that information last.

Exercise: Try the following two prompts using Copilot in Word and compare the responses you get. What differences do you see here?
- Prompt 1 (instruction-context-example): Craft a blog post outlining the benefits of meditation. Meditation is a centuries-old practice known for reducing stress. For instance, mindfulness meditation is known to provide positive effects on mental health.
- Prompt 2 (context-example-instruction): Meditation is a centuries-old practice known for reducing stress. Mindfulness meditation, for example, is known to provide positive effects on mental health. Given this information, craft a blog post outlining the benefits of meditation.
Tip: Iterate and regenerate. Often the first result isn’t the final or best response. If you’re not satisfied with the results Copilot gave you, revise your prompt, and try again.

Exercise: Try the following three iterations using Copilot in Word. See how the responses can improve with each iteration.
- Prompt 1: Write about advancements in renewable energy technologies.
- Prompt 2: Write a technical article that discusses recent advancements in renewable energy technologies, focusing on solar and wind energy. The target audience for this article consists of engineers and environmental scientists.
- Prompt 3: Compose a detailed technical article for engineers and environmental scientists. It should explore recent advancements and innovations in renewable energy technologies, with a focus on solar and wind energy. Discuss the scientific principles behind these technologies, their efficiency improvements, and real-world applications. Refer to recent research papers, patents, and industry reports for accurate and up-to-date information.

Continue