Taming the Copilot odyssey: Five ways to effectively prompt AI without going overboard

At Microsoft, we’ve been on our journey with AI and Microsoft 365 Copilot for a while now. We’ve made amazing strides and are still discovering new capabilities that are changing the way our employees work.

But is it possible that we’ve fallen in love with the journey, while forgetting our destination?

It’s easy to do.

When we turn to AI for help with a work task, we should always keep in mind our ultimate directive: How can we get the most out of these powerful tools with the least time, effort, and credits used?

In the epic story of “The Odyssey,” Odysseus spent 10 years trying to get home. Meanwhile, you spent the last hour trying to get what you needed out of Copilot.

Honestly, neither journey needed to be that dramatic.

We’re here to provide a reliable map that can help you avoid common AI pitfalls and make it back to Ithaca (or at least to your next meeting on time).

Don’t get sidetracked

It always starts with a quick task, right? “Summarize this meeting.” “Pull out the action items.” “Help me make sense of this thread before I crash out.” Then the adventure begins.

The first response you get is close, but not quite right, so you regenerate it: “Make it shorter.” Then you add more context, just because you can. Whoops, too much context! Now you’re wondering if you need to bring in another model or agent.

We know Copilot can help us get there, so we keep tweaking, keep trying one more thing. Sometimes that brings us closer to our goal. And sometimes it just turns a simple task into a whole odyssey.

Think of it this way: You’re fighting a series of beasts, and each one shines a light on prompting behaviors we need to leave behind. These detours are avoidable! You’re not only wasting time and credits; the effects of our AI habits on the environment are something everyone should be thinking about.

These five tips will help you get better results from Copilot faster, reducing unnecessary follow-ups and allowing you to travel fewer digital miles while you’re at it.

1. Write good prompts, not Greek tragedies

You write and send your prompt. Wait, you need to add the audience. Then you clarify the format. Oops, forgot to add the tone. Then you tell Copilot, “Also make sure it includes the thing from the meeting.” Each little follow-up feels harmless, but essentially you’re feeding your LLM the task one sailor at a time.

How to fix it: Remember Prompting 101!

Go back to the basics:

• Start with one clear prompt first
• Include the details you already know, before Copilot starts working: your goal, audience, format, tone, and source material 
• Clearly label specific instructions up front, so Copilot spends less time guessing

This way your future self won’t be stuck iterating (“shorter,” “try again,” “rewrite this part”). The goal isn’t a longer prompt—it’s a clearer one. Sail through once with the important pieces intact.

2. Tell it how long your voyage should be

AI has a habit of giving you long-winded answers every time you ask a simple question. Because you didn’t say “Keep this short” or “Give me a tidy bulleted list,” it gives you the whole weather system—every thoughtful little angle. Nobody has time for that.

Keep the bag of winds closed: Tell Copilot the form you want your answer in before it starts writing. For example:

• “Give me three bullets.”
• “Keep it under 100 words.”
• “Reply with a short Teams message.”
• “Make this a table.”
• “Give me the headline only.”
• “No background, just the decision and next step.”

The clearer you are up front, the fewer rewrites you’ll have to do later. Less is more, especially when AI is feeling chatty.

3. Steer clear of the context vortex

This beast shows up when you feed Copilot everything, “just in case.”

You’re working on a document, so you paste the entire thing into Copilot, when you really just wanted to change the title or tweak a few sentences. Now that you’re on a roll, maybe you add additional items to the chat: extra files, additional background, a meeting note, something loosely related but emotionally important. More. Lore.

Context can help, but more context is not always better. Sometimes you’re asking Copilot to swim through three thousand words so it can rewrite one sentence, which is not ideal.

How to stay afloat: Make sure Copilot is following the right current. Supply only the information it needs to complete the task, and avoid adding content that doesn’t need to be changed, summarized, compared, or otherwise used.

When you leave the unnecessary details behind, AI can work on the most relevant stuff without getting dragged under.

4. Pick the right boat before you set sail

Beware the siren’s song of using bigger and more powerful models because Why Not? They always promise you something very tempting: “Look at us—new and shiny! We’ll work better for you!”

Switching to a larger model feels more productive. But if you’ve got a simple task like summarizing or writing, you’re using way more money and energy than you need.

Before you jump on a ship, figure out what you really want: Do you need analysis? Research? Just a summary?

Decide on what “good” looks like ahead of time, then use the right Copilot experience for the job. Here’s a handy guide:

Ask yourself

5. Bring it on home (without taking a decade)

It sounds paradoxical: start at the end. Ask yourself, “Where do I want to end up?” Figure out your destination before you set off on your AI journey. Then work backwards to decide how best to get there.

All Rhodes should lead to your original goal. So, prior to setting sail, give Copilot a clear and specific destination. The sooner it knows where you’re going, the sooner you’ll get there.