
Your team does not have a prompt problem. It has a task problem.
Over the past year, many companies have sent people to prompt-writing workshops, shared lists of “perfect prompts,” and bought guides with names like “100 prompts for managers.” Then they waited for results. The results have been mixed at best.
The reason is simple, and you may already sense it. A prompt is a set of instructions. If you cannot explain a task clearly to a new colleague, you cannot explain it to a tool either.
The new hire test
Try this. Take a task you want AI to help with, such as replying to customer complaints. Write down the instructions you would give a new hire who starts on Monday. Then ask: could they do the job well using only those instructions?
Most people find they cannot. The writing is not the hard part. The answers are. What counts as a complaint worth escalating? When do we offer a refund? Who approves it? What tone do we use with a customer who has been with us for ten years?
Those are decisions about the work, not about the prompt. A better prompt cannot answer them. Only your team can.
What a defined task looks like
A task is defined when five things are written down.
1. The input. What arrives, in what form, and from whom.
2. The output. What a finished result looks like, with one real example.
3. The rules. What to do in the common cases, and in the edge cases you already know about.
4. The check. Who reviews the result, and what they look for.
5. The hand-off. Where the result goes next, and who needs to know.
When these five are clear, the prompt almost writes itself. You are copying answers into words. When they are not clear, no amount of clever wording will help, because the tool is guessing at the same gaps your new hire would.
What it looks like in numbers
Picture a team that spends 10 minutes on each first reply to a customer email, and sends 60 a week. That is 600 minutes, or 10 hours. These numbers are illustrative, so replace them with your own.
The team tries AI with a short, general prompt.
• Draft: 1 minute. Editing a weak draft: 8 minutes. Total: 60 × 9 = 540 minutes, or 9 hours
• Time saved: about 1 hour a week. Barely worth the change
Then the team spends one hour defining the task. They write down the rules, add three examples of good replies, and name who checks the results.
• Draft: 1 minute. Editing a good draft: 3 minutes. Total: 60 × 4 = 240 minutes, or 4 hours
• Time saved: 6 hours a week, or about 300 hours a year
The tool was the same in both cases. The difference was the clarity of the task.
Why the prompt myth is so popular
Prompt tips are easy to share. A good prompt fits on a slide. Defining a task means sitting with the people who do it, and often finding out that two of them do it differently.
That second part is uncomfortable. It is also where most of the value is. If two people do the same task in two ways, your company has no standard, with or without AI.
Fixing that is not a technology project. It is a conversation, and it usually pays off even if you never use a tool. New hires learn faster. Mistakes drop. Handovers get easier.
A short example of a missing rule
Imagine the AI drafts a reply to an angry customer and offers a full refund. The reply is polite and well written. It is also wrong, because your company only refunds after the item is returned.
No change of wording would have prevented that. The rule was never written down. It lived in the heads of two experienced staff members, who fixed the draft by hand every time, and never said why.
Once the rule is on the page, the tool follows it every time. And the next new hire follows it too.
Three objections you will hear
“We do not have time to write it all down.” You are already spending the time, in editing. Writing the page once costs about an hour. Editing weak drafts costs hours every week.
“Every case is different.” Some are. Most are not. Define the common cases first, and send the rest to a person. That alone covers the bulk of the work.
“Our best people just know how to do it.” That is the best reason to write it down. Knowledge kept in one head is a risk. Knowledge on a page is a shared asset.
Where prompts still matter
Prompts are not useless. Once the task is clear, a good prompt saves time and gives steadier results. Plain language works best: say what you want, give the rules, and show an example of a good result.
Keep the finished prompt in one shared place, with a name and an owner. When the task changes, update it. A prompt that lives in one person’s chat history leaves the company when that person does.
Also keep a short list of replies that needed heavy editing, and why. Those are the best clues for improving the rules. Most of the time, the fix is a missing rule, not a missing word.
How to define a task in one hour
Gather the two people who do the task most often. Ask each to complete three real examples while saying aloud why they make each choice. Write down every decision.
Then compare. Where the two differ, decide the standard together. Write one page: input, output, rules, check, hand-off.
You do not need a perfect document. You need a page that exists and that the team agrees with. You can improve it after the first month of use.
Where AI helps
AI can help you define the task, too. Give it your three examples and ask it to list the rules it sees in them. Then ask what is unclear. The questions it asks are often the same ones a new hire would ask.
Treat the answers as a first draft. Your team knows the exceptions, the customers, and the history. The tool can only work from what you show it.
One caution. Do not paste customer details or private company files into a tool until your company has a clear rule about what it may see. Use made-up examples while you are still defining the task.
Try it this month
1. Choose one task where your AI results have been disappointing.
2. Run the new hire test and mark the place where your instructions run out.
3. Fill in the five parts: input, output, rules, check, hand-off.
4. Turn the page into a prompt, and compare the editing time before and after.
Time the editing, not the drafting. The draft is always fast. The editing is where the real cost hides.
If the results are still weak after this, the page tells you where to look. That alone is worth the hour.
A better prompt starts with a clearer task.
Which task on your team would you struggle to explain to a new hire on one page?
If you want to talk through how to define a task before handing it to AI, I’d be glad to hear your thoughts.
ADDITIONAL READING
• You Trained Your Team on AI. Their Output Didn’t Change. — https://jordanimutan.com/2026/09/15/you-trained-your-team-on-ai-their-output-didnt-change/
• Not Every Task Should Use AI. Here’s the Test We Use to Tell. — https://jordanimutan.com/2026/09/25/not-every-task-should-use-ai-heres-the-test-we-use-to-tell/
• The 20% of Any Job That’s Actually Worth Giving to AI — https://jordanimutan.com/2026/10/06/the-20-of-any-job-thats-actually-worth-giving-to-ai/
• Your First AI Project Should Be Boring. Here Is How to Pick It. — https://jordanimutan.com/2026/11/07/your-first-ai-project-should-be-boring-here-is-how-to-pick-it/ • Why Your Next AI Win Doesn’t Need a Developer — https://jordanimutan.com/2026/10/08/why-your-next-ai-win-doesnt-need-a-developer/








