Most of what gets called prompt engineering is a handful of habits, and leaders need them more than engineers do. You are the one asking the tool for the analysis, the draft, the summary, the plan. If the question is sloppy, the answer is confident and wrong. These five habits fix most of it, and none of them needs a technical background. I teach them in a 3-hour course, and the delegates who get the most out of it are usually the least technical people in the room.
Here they are, with a before and after for each.
1. Give it the job, the audience and the constraint
The single biggest improvement is context. The tool cannot read the room. You have to describe it. Say who the output is for, what it is for, and what limits apply.
Before. ‘Write a summary of this report.’
After. ‘Summarise this board paper for a non-executive director who has 5 minutes, in under 200 words, in plain English, and end with the one decision that needs to be made at the meeting.’
The first gets you something generic. The second gets you something usable, because every clause in it removed a guess the tool would otherwise have made badly. Role, audience, length, tone, purpose. Five pieces of context, 30 seconds to type, and the difference between a draft you bin and a draft you send.
2. Show it an example
If you have a format you like, paste one in. A past report, an email that landed well, a table laid out the way you want it. The tool copies shape far more reliably than it follows a description of shape.
Before. ‘Write a client update in a professional but friendly tone with a clear structure.’
After. ‘Here is a client update I sent last month that worked well. [paste]. Write one in the same structure and tone for this month, covering these three points. [list]’
This is the fastest way to get output that looks like it came from your business rather than from the internet. One good example beats three paragraphs of description every time, and it means the tool learns your house style rather than inventing one.
3. Ask for the reasoning before the answer
For anything that matters, ask the tool to set out its assumptions and working first, then give the recommendation. Two things happen. The answer gets better, because it has to think before it commits. And you can see where it went wrong, which is the only way to trust it on the next question.
Before. ‘Which of these three suppliers should we choose?’
After. ‘For each of these three suppliers, list the assumptions you are making and the main risks you can see. Then rank them and explain the ranking. Then give a recommendation.’
A conclusion with no working is a guess with good grammar. The second version shows you the guess, so you can correct the assumption it got wrong about your business before you act on the recommendation. Most bad AI advice comes from a hidden assumption, and this habit drags it into the light.
4. Treat the first answer as a draft
Leaders tend to take the first output or reject the whole thing. The productive habit is to push back. The conversation is the work, and a good tool improves sharply on the second and third pass.
Before. First draft arrives. ‘That is not quite right.’ Close the window.
After. ‘That is too long. Cut it to half. You have missed the cash flow risk, add a paragraph on it. Make the second option the recommendation and explain why. Now make the opening sentence state the decision.’
Four short instructions, two minutes, and the draft is close to done. The first output is where the tool shows you what it understood. The follow-ups are where you fix what it misunderstood. Stopping after one pass is leaving most of the value on the table.
5. Check the things it cannot know
AI tools are fluent about facts they do not have. Your numbers, your contracts, last week’s decision, the thing the client said on the call. Any output that touches those needs a human check against the source.
Before. The tool produces a proposal with a price, a delivery date and a reference to your terms. It goes out.
After. Before anything goes out under your name, ask ‘where did it get that’. The price came from where. The date is based on what. The terms it quoted, are they ours. Anything the tool could not have known from what you gave it is a guess, and the guesses are where the embarrassment lives.
Most of the public AI stories in business come from skipping this step. A fabricated citation, a wrong figure, a term that does not exist. The habit is cheap and it is the one that protects your reputation.
The habit behind the habits
Good prompting is clear thinking written down. If you cannot say what you want, who it is for and what good looks like, the tool cannot either. Leaders who are already precise with people get precise results from AI almost immediately. The rest learn precision quickly, because the tool punishes vagueness in a way colleagues are too polite to.
There is a useful side effect. Leaders who practise these habits with AI tend to brief their people better too, because the discipline is the same. Say what you want, for whom, within what limits, show an example, ask for the reasoning, iterate, check the facts. That was always good management. The tool just makes the cost of skipping it visible.
The common mistakes
Writing one long prompt and expecting a finished answer. Break it into a conversation.
Describing the format when you could paste an example. The example wins.
Trusting fluency. The better it reads, the more carefully you should check the facts.
Putting confidential data into tools nobody has approved. Bring the task, leave the data behind, and know which tools your organisation has cleared for which information.
Where to start
Take one task you do every week that involves writing or analysis, and rebuild your prompt for it using the five habits. Save the prompt. Use it next week. Improve it. Within a month you will have a small library of prompts that do real work, which is exactly what we build in the Practical Prompt Engineering course. An AI opportunity review is the right starting point if you want to find where AI earns its keep across the business first.
Frequently asked questions
Do I need to be technical to write good prompts?
No. The habits that matter are clarity, context and checking, which are leadership skills rather than technical ones. Technical knowledge helps at the edges, not at the core.
What is the most common prompting mistake?
Asking for an output without saying who it is for, what it is for and what limits apply. Context is the difference between a generic answer and a useful one.
Should I use one long prompt or several short ones?
Several. Give the context and the job, read the first draft, then refine with short follow-up instructions. The conversation produces far better results than a single long request.
How do I know if an AI answer is reliable?
Ask for the reasoning and the sources, then check anything that relies on facts the tool could not have, such as your own numbers or recent events. Trust the working, not the fluency.
Which AI tool should I use?
Whichever your organisation has approved. The five habits work the same in all of them. The tool matters far less than the quality of the instruction.
Is prompt engineering a passing skill?
The tools are getting better at understanding vague requests, but clear thinking about what you want never goes out of date. The habits transfer to every tool that follows.