Most people use AI like a better search engine: they type a short sentence and wait for a miracle. Yet all it takes is changing the way you ask, and the quality of the answers goes up. I'll show you seven techniques I regularly teach in my training sessions. These aren't academic concepts, but practical ways to change how AI thinks and get more out of it.
Article summary
- Role prompting: tell the AI who it should be.
- Context prompting: describe the situation, not just the task. This is the technique with the biggest impact.
- Few-shot prompting: show examples of what the result should look like.
- Chain prompting: break a complex problem down into steps.
- AI as a critic, opponent and consultant: let the AI look for weak spots and counterarguments and ask you questions.
In the article Prompt is king I wrote about why the prompt matters. This time we're going a step further. The following techniques work in ChatGPT, Microsoft Copilot, Claude, Gemini and other tools. You don't need any programming for them, just a slightly different way of thinking.
For each technique, you'll find a short explanation, a sample prompt you can copy right away, and a tip on when it's useful.
1. Role prompting: tell the AI who it should be
Instead of a general question, you give the AI a specific perspective. The same task is handled differently by "someone in general" than by an experienced manager, lawyer or marketer.
Without a role:
Write feedback for an employee.
With a role:
Act as an experienced team manager with 15 years of experience.
Write constructive feedback for an employee who consistently delivers quality work but misses deadlines.
When to use a role
Whenever you need the perspective of a specific profession, such as a manager, HR specialist, lawyer, project manager, data analyst, marketer or consultant.
What you gain
The AI can better judge the right level of detail, communication style and criteriaby which it should evaluate the result.
2. Context prompting: describe the situation, not just the task
This is a technique I recommend using always. The biggest difference in answer quality usually doesn't come from the role, but from the context. If you don't tell the AI what it's about, it has to guess, and the result will be generic.
Without context:
Write an email about a change of deadline.
With context:
I'm preparing an email for an important customer.
The project will be delayed by two weeks because of a late delivery from an external partner.
The customer has been satisfied for a long time and we want to keep a good relationship.
Write an email about a change of deadline.
What you gain
The AI doesn't have to guess the situation, audience or goal. Three extra sentences of context often save you five rounds of corrections.
A good prompt isn't a short one. A good prompt is one after which the AI doesn't have to guess.
3. Few-shot prompting: show examples
You don't just say what you want. You show what it should look like. AI is excellent at recognizing patterns, so two or three examples often work better than a paragraph of explanation.
Example:
Rewrite task names in this style:
Example 1:
"We'll do the website" → "Prepare a draft of the website structure"
Example 2:
"Take a look at the offer" → "Evaluate the supplier's offer"
Now rewrite:
"Deal with onboarding"
When to use examples
- when sorting and labeling data,
- when rewriting texts into a consistent style,
- when you want to follow your company's communication style,
- when you can't put into words exactly what you want.
4. Chain prompting: break the problem down into steps
A complex task given in one sentence leads to a superficial answer. When you break it down into steps, the AI proceeds more logically and you can check and correct each intermediate result.
Instead of:
Come up with an AI adoption strategy.
Try:
1. Analyze the company's situation.
2. Identify the barriers.
3. Suggest quick wins.
4. Suggest pilot projects.
5. Suggest an adoption plan.
A simpler version also works: "Ask questions first. Then analyze. Finally, suggest a solution."
When to use steps
For strategies, projects, analyses and decision-making. The result is less confusion and better logic.
5. AI as a critic: let it look for mistakes
AI is eager to help you, and that's why it tends to praise you. But when you explicitly ask it for criticism, you get a second opinion for free. You can even have it evaluate its own output.
After writing a presentation, add:
Evaluate your presentation.
Find:
– weak spots
– unclear points
– missing arguments
– unsupported claims
Or for your own document:
Read my project proposal.
Be critical.
Find reasons why the project could fail.
Don't hold back on criticism.
6. AI as an opponent: who will push back on you?
Similar to the critic, but the AI is given the role of a specific person who disagrees with you. The critic looks for mistakes in general; the opponent looks for them from a specific perspective.
Example:
Act as the chief financial officer.
Find reasons why this investment should not be approved.
Or:
Act as a skeptical customer.
What objections would you have?
When to use an opponent
It's great preparation before a sales meeting, a presentation to management, a workshop or an investment proposal. You'll hear the objections before a real person raises them.
7. AI as a consultant: ask for thinking, not output
Here the roles are reversed. You don't want a finished text, you want a sparring partner to help you think the problem through.
Wrong:
Write a marketing strategy.
Better:
Act as an experienced marketing consultant.
Don't jump to solutions.
First ask me the questions you would need answered in order to propose a marketing strategy.
And after your answers:
Based on my answers, suggest:
– options
– advantages
– risks
– recommendations
The AI will make you think. Often the greatest value isn't in the answers, but in the questions you wouldn't have come up with yourself.
One sentence to remember
| Technique | What it does |
|---|---|
| Role prompting | I tell the AI who it should be |
| Context prompting | I provide the situation and context |
| Few-shot prompting | I show examples |
| Chain prompting | I break the problem down into steps |
| AI as a critic | Looks for weak spots |
| AI as an opponent | Looks for counterarguments |
| AI as a consultant | Helps you think and asks questions |
If you only remember three
The biggest impact comes from context prompting, AI as a consultant and AI as an opponent. These three techniques are exactly the ones most regular users hardly ever use, yet they raise the quality of outputs more than any complex prompt engineering tricks.
The techniques can also be combined. Give the AI a role, describe the context, ask it to ask questions first, and at the end let it tear your proposal apart as an opponent. That's already a workflow, not just a question in a chat.
Frequently asked questions
What is a prompt?
A prompt is the instruction you write to an artificial intelligence, such as ChatGPT or Microsoft Copilot. It can be a question, a task or an instruction. The quality of the answer depends heavily on how you write the prompt.
Which prompting technique is the most important?
Context prompting, i.e. describing the situation, the audience and the goal. The biggest difference in answer quality usually doesn't come from the role, but from the context. I recommend using it every time.
What is the difference between AI as a critic and AI as an opponent?
The critic looks for mistakes in general, such as weak spots, unclear points or unsupported claims. The opponent is given the role of a specific person, such as a CFO or a skeptical customer, and looks for counterarguments from their perspective.
Do these techniques work in both ChatGPT and Microsoft Copilot?
Yes. The techniques described aren't tied to a specific tool. They work in ChatGPT, Microsoft Copilot, Claude, Gemini and other language models.
What is few-shot prompting?
Few-shot prompting means you show the AI a few examples in your prompt of what the result should look like. The AI picks up the pattern and continues in the same style. This often works better than a long explanation.
Does a good prompt have to be long?
Not necessarily. A good prompt is one after which the AI doesn't have to guess the situation, the audience or the goal. Adding two or three sentences of context is usually enough.
Do you want your team to be able to work with AI in a practical way, without unnecessary technical jargon?
Take a look at my AI courses or follow more practical guides on my blog.