A friend of mine tried Copilot once and decided it wasn’t for her. She asked it to write a message for her book club, a “here’s what I thought” note about the novel they were reading. What came back was three tidy paragraphs of what she called “book report language.” Nothing was wrong. Nothing was hers.
She closed the tab.
I asked what she’d typed. “Write me a message about the book.”
That’s the whole story. The instruction was thin, so the response was generic. Copilot has no way to know you want casual when you say nothing about it, so it fills the gap with what usually works, which is formal and complete, like a book report.
She tried again a few weeks later. This time she gave it more: who the note was for and the tone she wanted, plus a bit of her own personality. Still not perfect on the first try. Still fixable. She edited it in three minutes and sent it. Her book club thought her writing had improved.
Here’s what changed: the tool did the exact same thing both times. What changed was how much she told it.
What this teaches you
Copilot doesn’t try to nail it on the first try. It generates a plausible first draft from whatever you gave it, and thin instruction produces generic output. That is how the tool works, and the real work happens in the back-and-forth.
If the first draft doesn’t fit, you stay in the same chat and tell it what to change. It keeps everything you’ve said in that conversation in front of it, so by the second or third version it’s working from all of your corrections, not just the last one. Two or three rounds is usually enough.
Try this
Pick something you tried before that came out wrong, or start fresh:
- Open Copilot (copilot.microsoft.com or your Copilot-enabled browser) and type your request with context: who it’s for, what you want it to say, and the tone you’re going for.
- Read the draft. It probably won’t be exactly right, and that’s normal.
- Stay in the same chat and tell it what you want changed. Be specific: “Can you make it shorter?” “It’s too formal. Can you write it the way I actually talk?” “Can you end on the part about the garden instead?”
- Read the second version. Do it again if you need to.
Most of the time, two rounds gets you close enough that you’re only fixing a sentence or two yourself. Each correction builds on the last one, so by round three it has a much better idea of what you actually want.
What to watch for
In my experience, people stop too soon. If something is 80% right and you’re not sure how to fix it, describe what’s wrong: “It’s almost there, but something feels off about the opening.” You don’t need to know the answer. Naming the problem is enough.
One real limit: if you’re asking for something specific, a fact or a date, you still have to check the answer. Copilot sounds confident even when it shouldn’t be. It’s genuinely good at sounding right, and you’re the one who knows what’s actually true. Stay skeptical on facts.
And sometimes iteration just doesn’t work. If you’ve done five rounds and it’s still missing the mark, you might be asking for something the tool can’t do, or you might just need to write it yourself. Some tasks are a poor fit for the back-and-forth, and you’ll learn which ones as you go.
My friend now uses Copilot for every book club message. The first draft is still never exactly right, and she’s made peace with that. The second one usually is. She told me she’s started looking forward to the writing part.
Once you’re comfortable with iteration, the next skill in this series gets easier. You’ll have a feel for when to push back and when to accept, and for how much context to give up front. The tool works better every time you use it, because you get better at telling it what you want.
Making Sense of Artificial intelligence (AI) is a six-part series walking first-time users through the basics, one skill at a time.