Five tips for prompting AI: How we’re communicating better at Microsoft with Microsoft 365 Copilot

It’s all in the prompt.

The AI revolution is here, and we’re finding that the best way to take advantage of it is to learn how to prompt it.

Here in Microsoft Digital, the company’s IT organization, we’ve had early access to Microsoft 365 Copilot. We’ve learned, among other things, that the better your prompt, the better help you get.

Our internal IT communications team was one of the teams that put Copilot to work right away, and they spent a lot of time figuring out what prompts work the best.

“We looked for ways that Copilot could help us communicate with employees more effectively,” says Sarah Lundy, a senior program manager on the IT Communications team in Microsoft Digital. “We quickly realized that it was all about prompting Copilot—the better your prompt, the better your result.”

Here are five things Lundy and her team learned about prompting Copilot.

{Learn more about how we’re deploying Microsoft 365 Copilot internally at Microsoft. Unpack how we’re getting the most out of generative AI at Microsoft with good governance. Explore how we’re using Microsoft 365 to bolster our teamwork.}

Getting your prompt on

Start by letting Copilot know what role you want it to play, so it can couch its answers appropriately. For example, you could tell it, “You’re an internal communicator, please craft a message that lets employees know that they need to run our newest Windows security update.”  

“We all have particular jargon or a specific way of speaking,” says Laura Oxford, a senior program manager of IT communications and a beta tester of Copilot. “Give Copilot a persona so it can generate responses from that perspective.”

Simply put, the objective is what you’d like Copilot to do. Summarize a meeting? Draft an email? Create an executive summary of a presentation?

“It doesn’t have to be formal,” Oxford says. “The objective can be something loose like, ‘brainstorm some ideas for a document.’ I often just brain-dump with it. I’ll have all these thoughts swirling in my head, and they’re not quite formulated. It can feel intimidating to look at a blank page and not know how to fill it up. This gives me something to react to. It can be a really good starting place.”

Knowing your audience and making that clear to Copilot is an essential element of prompting, because different audiences have different needs and expectations. Are you trying to reach a manager with 20 years of experience? A new entry-level employee? An IT pro? A busy CEO?

“Tell that to the tool,” Lundy says. “We’ve found it performs better depending on whether you tell it stuff like it’s an internal email celebrating a win versus an external communication to customers.” 

Whether you want to sound informal, professional, or somewhere in between, Copilot can help you best when it understands what you’re looking for in terms of tone, length, word count, or any other limitations or boundaries you set.

For example, a highly technical, hour-long presentation at a conference will have a different tone than a tweet. A humorous tone is probably not appropriate for a message to an executive. Conversely, a formal tone might not be the best idea for a social media post.

Here again, the quality of the output is contingent upon the quality of the input. Tell Copilot whether you expect a 6th-grade reading level or the level required to understand an academic paper. A summary meant for someone in the C-suite will have different parameters from a consumer-facing blog post. Tell it how many words, what format you’d like it to turn out (such as a bulleted list), and whether there are any restrictions, then let it do its thing.

Context covers a wide range of potential Copilot inputs, so be as specific as possible and add as much information as you can. Generally, the more context, details, and info you provide, the better your results will be. If you have an example document similar to the format you’re going for, show that to the tool. You’ll be surprised at how much better the results are when you give Copilot clear instructions.

That said, you can keep it casual.

“People tend to skimp on context because they don’t necessarily want to put in the work, and they kind of expect the tool to read their minds,” Lundy says. “You might see that the content is great, but that it’s not exactly what you wanted. If you go a little deeper and add something like, ‘Make sure you touch on these three items,’ you can get dramatically better results, which can help you avoid having to make edits later.”

You don’t necessarily have to write a long prompt for it to be effective.

“Give it some bullet points so it knows about those things that you already know in your mind you want to be included, even if you haven’t completely thought it through,” Lundy says.

She adds that prompting gets faster as you go. She advises saving the prompts that work really well to customize and reuse later.

Iterating and improving

We learn more every day about generative AI’s capabilities and our ability to interact with it. One of the main lessons from our Customer Zero testing is that prompting is an art, not a science, and it requires experimentation to see how you can get the most out of it for your own purposes.

“Play around with it. Have fun with it. You can’t break it, so just experiment,” Oxford says.

In addition to helping her brainstorm and get past the dreaded blank page, Oxford uses it to help her offload mundane tasks so she can focus on more productive, creative, and high-level work.

 Using a smart framework for prompts can help you. “You don’t need any technical skills,” Oxford says. “You just have to help it figure out how it can help you.”

We learn more by the day about best practices for using this incredible tool, and everything we learn, we pour back into Copilot using our Customer Zero framework—and the more we discover, the better we make the product for our users.