A Practical Guide to Prompting AI

Most people know how to ask AI a question. But asking AI a question and prompting AI effectively are two different things.

If you've ever asked ChatGPT or Claude something and thought "that's not what I meant", the problem isn't necessarily the AI. Often, the problem is that we haven't explained clearly enough what we actually want.

The good news is that you don't need to learn complicated prompt engineering techniques or memorise magic phrases. You just need to communicate clearly.

First: what is a prompt?

A prompt is simply the instruction or information you give an AI. For example:

Write a LinkedIn post about artificial intelligence.

That's a prompt. But it's a very broad one. The AI doesn't know:

• Who the audience is
• What your opinion is
• How long the post should be
• Whether you want a serious or casual tone
• Whether you're trying to educate, entertain or sell something

So it has to make assumptions. A better prompt would be:

Write a LinkedIn post for business owners who are new to AI. Explain in simple language how AI can save time in a business. Keep it conversational, practical and under 500 words. Don't make it sound like a sales pitch.

The more clearly you communicate your goal and relevant constraints, the better the AI can usually respond.

The simple formula for a good prompt

You don't need every prompt to be long. But for more important tasks, think about these five things.

1. Context

Give the AI the background information it needs. Instead of "help me fix this", try: "I'm running a Ruby on Rails API on AWS using two EC2 instances behind an Application Load Balancer." Now the AI understands the environment.

Ask yourself: what does the AI need to know before it can give me a useful answer?

2. Task

Tell the AI exactly what you want it to do. Instead of "look at this code", try "review this code and identify the cause of the error", or "review this code for security vulnerabilities", or "refactor this code to make it easier to maintain".

"Look at this" is vague. "Identify the cause of the error" gives the AI a clear job. Use direct instructions such as write, explain, analyse, compare, summarise, review, debug, rewrite and create.

Clear, explicit instructions are a core recommendation in both OpenAI's and Anthropic's prompting guidance.

3. Constraints

Tell the AI about any important rules. For example:

• "Explain this to a non-technical person."
• "Keep the answer under 300 words."
• "Don't change the database structure."
• "Use South African English."
• "Give me practical examples."
• "Don't guess. If you aren't sure, say so."

Constraints can be incredibly useful because they prevent the AI from making assumptions you don't want it to make. Modern prompting guidance specifically recommends defining relevant context, requirements and the desired level of detail.

4. Expected output

Tell the AI what you want the answer to look like:

• "Give me the answer as a table."
• "Give me a step-by-step guide."
• "Give me five bullet points."
• "Write this as a LinkedIn post."
• "Return only the JSON."

Don't assume the AI knows what format you want — tell it. For particularly specific formats, showing the AI an example of the desired structure can improve consistency.

5. Give it the right information

This is probably one of the most important parts of prompting. AI is not magic. If you're asking it to analyse a problem, give it the information needed to analyse that problem.

Instead of "my website is slow, what's wrong?", you could say:

My website is hosted on AWS. The application runs on two EC2 instances behind an Application Load Balancer. The database is MySQL RDS. Pages take approximately 5 seconds to load. The CPU usage on the EC2 instances is below 20%. Help me identify the most likely causes and give me a step-by-step troubleshooting plan.

Now the AI has something useful to work with.

The biggest prompting mistake: being too vague

"Write me a marketing plan" will probably produce a generic answer. But this:

Create a 90-day marketing plan for a small South African software development company. Our target audience is small and medium-sized businesses that need custom web and mobile applications. We have a limited marketing budget. Focus on LinkedIn, content marketing and referrals. Give me a week-by-week action plan.

is much more likely to produce something useful. The AI can't read your mind. If something is important to the answer, tell it.

Do you need to say "please" and "thank you"?

No. Technically, extra words mean extra text for the model to process. In API-based AI systems, usage is commonly measured in tokens, which are chunks of text used by the model.

But don't get obsessed with removing every "please" and "thank you". For most people, those few words make virtually no meaningful difference. The bigger problem is unnecessary context. For example, don't give the AI:

• 50 pages of documentation when one page is relevant
• Your entire email history when you need help with one email
• A massive codebase when the problem is in one function
• Five previous versions of something when only the latest version matters

Don't optimise politeness. Optimise relevance.

More information isn't always better

People sometimes think the more information you give AI, the better the answer will be. Not necessarily. Imagine asking "why is my application returning a 500 error?" and then pasting your entire application log, your complete database schema, 20 configuration files, three years of deployment history and every environment variable.

Some of that information may be useful. Most of it may not be. A better approach is to start with the information most relevant to the problem. Modern AI guidance increasingly emphasises the quality and relevance of context, rather than simply putting as much information as possible into the conversation.

Tell AI who the answer is for

The same explanation can be completely different depending on the audience. Compare "explain AWS" with "explain AWS to a business owner who has no technical knowledge" and "explain AWS to a junior DevOps engineer preparing for an interview".

Same topic. Very different answers. Always consider who is going to read or use the answer.

Ask for examples

If you're learning something, examples are often more useful than explanations. Instead of "explain what an API is", try "explain what an API is in simple language — use a real-world example and then show me a simple technical example". Examples are also useful when you want the AI to follow a particular format or writing style.

Don't ask AI to guess

This is especially important when accuracy matters. Instead of "analyse this information", try:

Analyse this information. If there isn't enough information to reach a conclusion, tell me what information is missing. Do not make assumptions or invent information.

Giving the AI permission to say it doesn't know can help reduce confident guesses when the available information is insufficient.

One big task vs smaller tasks

Sometimes people give AI an enormous instruction: "analyse my business, review my competitors, create a strategy, write a marketing plan, create 20 LinkedIn posts and build me a website". That's a lot. You may get better results by breaking it into smaller tasks:

• Step 1: "Help me define my target audience."
• Step 2: "Based on the target audience we defined, identify their biggest problems."
• Step 3: "Based on those problems, create five content themes for LinkedIn."
• Step 4: "Write a LinkedIn post based on content theme number three."

Breaking complex work into smaller, focused tasks can make it easier to review and refine each stage.

Don't expect the first answer to be perfect

One of the biggest misconceptions about AI is that you need to write the perfect prompt the first time. You don't. Think of AI as a conversation. You can say "that's too technical, explain it more simply", or "I like the first half, but make the second half more practical", or "make this sound less like marketing", or "give me three alternatives", or "keep the same content, but make it 50% shorter".

Prompting is often an iterative process: review the result, identify what's missing, and refine the request.

A simple prompt template

When you're not sure how to ask, use this:

• Context: here's what you need to know…
• Task: here's what I want you to do…
• Constraints: here are the important rules…
• Output: this is what I want the answer to look like…

Context: I'm writing LinkedIn content about AI for business owners who are not technical. Task: write a post explaining what AI tokens are. Constraints: use simple language, avoid technical jargon, make it conversational, don't make it sound like an advertisement, keep it under 500 words. Output: start with a strong hook, explain the concept with a simple analogy, and finish with a question that encourages discussion.

That's a clear brief. And clear briefs generally produce better results.

Common prompting gotchas

Being vague

"Help me with my website." Better: "My website takes more than five seconds to load on mobile. Help me identify the likely causes and give me a prioritised list of things to investigate."

Asking for too many things at once

"Build my entire business strategy." Better: break the problem into smaller pieces.

Forgetting the audience

"Write an article about cloud computing." Better: "Write an article about cloud computing for small business owners with no technical background."

Not specifying the format

"Analyse this data." Better: "Analyse this data and return the top five findings in a table with columns for finding, evidence and recommendation."

Giving irrelevant information

More information isn't automatically better. Give the AI the information that matters.

Treating AI as a search engine that always knows the answer

AI can be wrong. It can misunderstand information, make incorrect assumptions, and sometimes present incorrect information confidently. For important decisions, verify the answer using reliable sources and your own expertise.

Using complicated "magic prompts"

You've probably seen prompts that start with "you are the world's greatest expert with 30 years of experience, think like a genius, analyse this using first principles…". Stop. Modern AI models generally don't require elaborate role-playing or magic phrases for ordinary tasks. Start with a clear request and add detail only when it helps.

The most important lesson

Good prompting is not about knowing secret commands. It's not about saying "act as an expert". It's not about finding a magical prompt on the internet. And it's definitely not about whether you say "please".

Good prompting is simply about communicating clearly. Give the AI:

• The information it needs
• A clear task
• Relevant constraints
• The format you want
• An opportunity to ask questions or admit uncertainty when appropriate

Then review the answer. And refine it.

The best AI users aren't necessarily the people who know the most about artificial intelligence. They're often the people who know what they want to achieve and how to explain it clearly.

Don't focus on finding the perfect prompt. Focus on being clear.