Stop asking AI questions. Start delegating to it.
Most people use ChatGPT or Copilot at work the way they use Google. Type a question, read the answer, decide whether to trust it.
That habit is the reason so many people give up on it. A search engine shows you sources and leaves the judgement to you. An AI tool gives you one fluent, confident answer, and nothing on the page tells you whether it's right. Ask it a question with no context and it fills the gaps with guesses. Then it presents the guesses in exactly the same tone as the facts.
So people conclude that "AI makes things up" and go back to doing the work themselves. That's a fair reaction to the results. But the problem is how it's being used, not the tool.
A question versus a delegation
Here's the difference, using a job most finance and admin teams know.
Asking a question:
Write a summary of our sales for this week.
The AI doesn't have your sales figures. It doesn't know who reads the summary, what "this week" covers, or what last week's looked like. So it either asks you for all of that or, worse, invents something plausible.
Delegating a job:
You're helping me prepare the weekly sales summary for the leadership team. They read it on Monday morning and want to know in two minutes whether we're on track.
Attached are this week's figures and last week's summary. Use the same structure as last week: headline numbers, three bullet points on what changed and why, then anything that needs a decision.
Only use figures from the attachment. If a number is missing or looks odd, don't estimate it. List it under "Questions for me" at the end.
Keep it under 200 words, in plain English, with no jargon.
Same tool, same job. The second version gives the AI what any new starter would need, and it tells it what to do when it isn't sure. That last part is what stops most of the "it made it up" moments.
The six-part brief
You don't need special prompt tricks. A good brief answers the six questions you'd answer for a trainee on their first week:
- Who is this for, and why? The audience and what they need from it. "Leadership, on Monday, wants to know if we're on track."
- What's the job? One clear task, not three. "Prepare the weekly sales summary."
- What do you have to work with? The inputs: attached figures, last week's version, your template. If the AI can't see it, it can't use it.
- What should it look like? Format, length, structure, tone. An example of a good one beats a paragraph of description.
- What's off limits? "Only use figures from the attachment." "Don't include customer names." Boundaries prevent the confident guess.
- What happens when it's unsure? Tell it to flag, not fill. A list of questions at the end is far more useful than a smooth paragraph built on an assumption.
Write that once for a job you do every week, save it, and reuse it. The first time takes ten minutes. After that it takes seconds, and the output gets more consistent each week because the brief stays the same.
You still check the work
Delegating isn't abdicating. You'd check a trainee's first few reports line by line, and you should check the AI's too. The difference is that now you're reviewing a draft instead of building one from scratch. That's where the time comes back.
As the brief improves, the checks get quicker. You learn where it tends to slip, and you add a line to the brief to cover it. That loop of brief, check and improve the brief is most of what good AI training actually teaches.
Try it on one job this week
Pick one task you do every week that you could hand to a trainee with a checklist. Write the six-part brief for it. Run it, check it, and adjust the brief.
If you're not sure which job to start with, or what's been holding you back so far, the readiness check below takes about three minutes. It tells you which of the four Ds to work on first: Delegation, Description, Discernment or Diligence.