Someone stopped me in a corridor and asked for “a quick one-pager on the options”. I said yes, went back to my desk and started writing.
Two days later I sent it over. The reply was polite and deflating: “Thanks, but that’s not really what I meant.” They wanted a recommendation, not a list of options. And they wanted it for a meeting with people I didn’t know were involved.
The writing was fine. The problem was that I’d started before I knew what the ask actually was. These days I spend ten minutes getting clear first, with AI doing the bit I’m worst at: spotting what’s missing. Here’s how.
Step 1: Capture the request exactly as it came
Write down the ask word for word, plus anything you know about it. Who asked, when they need it, and where it’s going (a meeting, an email, a decision). Don’t tidy it up. The vague bits are the useful bits right now.
If it came by email or chat, copy it in. For work requests, only use the AI tool your company has approved, and leave out anything confidential you wouldn’t store outside company systems.
Step 2: Ask AI to find the gaps, not do the work
The temptation is to ask AI to write the thing straight away. Don’t, yet. Ask it to read the request like a sceptical colleague would.
Here’s the prompt I use:
“Here is a request I’ve been given, plus what I know about it. Don’t do the task. Instead: 1) List anything that is unclear or could reasonably mean two different things. 2) List what I’d need to know to do this well, such as audience, purpose, format, length, deadline and who makes the final call. 3) Give me the five most useful questions to ask the person before I start, phrased in a friendly, short way. Don’t invent details about the person or the situation.”
You’ll usually get more questions than you need. That’s fine. The point is to see the gaps laid out in front of you.
Step 3: Pick two or three questions, not ten
Nobody wants a questionnaire back. Pick the ones where a different answer would change what you make. In my experience they’re nearly always some version of these:
- What’s it for? A decision, an update, or something to forward on?
- Who’s it for? Just them, or a wider group you haven’t met?
- What does good look like? A recommendation, a list of options, a rough first pass?
Deadline and format matter too, but people often tell you those without being asked.
Step 4: Send them in your own words
Rewrite the questions so they sound like you. Keep it light. Something like: “Happy to. Quick check before I start so I get it right: is this to help you decide, or to share with the wider team? And would a recommendation be more useful than a list of options?”
That takes thirty seconds to read and saves you from a two-day detour. You send it, not the tool.
Step 5: Turn the answers into a short brief
Once you have the answers, paste them back in and ask AI for a five-line brief: purpose, audience, format, deadline, what good looks like. Check it, fix anything it got wrong, and keep it next to you while you work. If the request drifts later, you’ve got something to point back to.
Where AI stops
AI is good at spotting what a request doesn’t say. It doesn’t know the person asking, the history behind the request, or the politics of who else cares about the answer. Sometimes the right move is just to walk over and ask in person. Use AI to find the gaps, then use your judgement about which ones are worth asking and how.
Quick questions
Won't asking questions make me look like I wasn't listening?
Not if you keep it to two or three short, friendly questions. Most people are glad you checked before spending days on the wrong thing.
Should I let AI do the task once I have the answers?
It can help with a first draft, but check and rewrite it yourself. The value here is getting clear on the ask, not handing over the work.
Is this okay for work requests?
Only use the AI tool your company has approved, and leave out anything confidential you would not store outside company systems.