The art of prompting: here's how to write the perfect prompt in 10 steps

An AI tool can do a lot, but without proper instructions, it works mostly on assumptions. Not so useful with content that needs to convert. Do you want AI to feel like a smart colleague working with you rather than a text generator just trying something out? Then you need to learn how to write good prompts. With the right instructions, not only will you get better results, but you'll save time and ensure consistent, relevant content. In this article, we'll dive into the world of prompting and show you how to effectively drive AI for the best output.

What is prompting?

With prompting, you give an AI system clear instructions or questions to arrive at the desired output. You can do this directly in a chat message, but you can also think of the instructions in your custom GPT or our AI content assistant as a prompt. It's more than just asking a question: it's about giving the right context, details and expectations so that the AI understands exactly what you mean and can give you what you need.

Think of prompting as breaking in a new colleague. This colleague has the knowledge and experience, but does not yet know how your processes work or exactly what is expected of him or her. You must also first induct this colleague, train him or her with handbooks and manuals, and introduce him or her to the company and colleagues. Just as you cannot expect a new colleague to do everything perfectly right away, you cannot expect AI to do the same.

Prompting thus forms the basis of how you work with an AI, whether it's answering questions, creating creative content, or solving complex problems.

Different types of prompting

You can use prompting in different ways. For content, we mainly work with three task types:

• Writing: AI creates textual output, such as blogs, social posts, emails, ads or landing pages.
• Analyzing: AI evaluates input, extracts insights from data or provides points of improvement, for example in SEO analysis, competition research or content audits.
• Combine: AI first analyzes input and then writes an improved version, for example when rewriting a landing page or optimizing a blog.

That first choice is important. A writing task requires tone of voice, style and format. An analysis task, on the other hand, requires criteria, sources and sharp validation. If you lump everything together, then you also get output in one heap. Cozy, but rarely useful.

Content prompting

Content prompting is the targeted use of AI for textual output, such as blogs, social media posts, emails, ads and landing pages. It is not about "just create a text," but about driving AI based on task, purpose, target audience, tone of voice, knowledge, format and conditions.

At Sterc, we don't use AI to just create more content. We use AI to build better content processes: faster from input to concept, more consistent in tone of voice and more sharply aligned with audience and purpose.

Content prompting helps you to:

• Maintain a consistent tone of voice, such as down-to-earth, advisory or activating.
• Match content to target audience, knowledge level and stage in the customer journey.
• Choose the right format, such as blog, social post, email, case or landing page.
• Pass along guidelines for SEO, brand style, sources, claims and CTAs.

Why is a good prompt so important?

So a prompt is really nothing more than the question or command you give to an AI tool. Sounds simple, right? Yet this is where things often go wrong. A bad prompt is like giving an unclear assignment to your team: you get a messy result. A good prompt, on the other hand? That one makes the AI work as if you were hiring a specialist.

An effective prompt usually contains several elements, such as:

• Context: Why are you asking this question, and what is the background information?
• Specific details: What exactly should the AI take or avoid?
• Purpose: What do you ultimately want to achieve with the answer?

By combining these elements, you give the AI all the tools it needs to respond in a targeted and relevant way.

For example, if you want to ask an AI to help write a blog about digital accessibility, you might choose to include not only the topic, but also details about the target audience, desired writing style, and specific topics. Only when these components are clear will the output exceed your expectations.

Let's get practical for a moment:

❌ Bad: "Write a text about marketing."

✅ Good: "Write a LinkedIn post for SME entrepreneurs on how to use social media smarter to get more inquiries. Write in Dutch, keep the tone practical and slightly enthusiastic, use a maximum of 120 words and end with a concrete question."

See the difference? It's all about clarity and context. AI is incredibly powerful, but it can't guess what you mean. You set the direction, and AI follows. The better we steer, the better the results.

Extra: what makes modern AI models different

Whereas with older AI models you could often still get away with a loose task, modern models require more clarity. Not because they are dumber, but because they follow instructions more precisely.

That's convenient, as long as you know what you're doing. If you give conflicting instructions, you'll also get conflicting output. If you ask for an expert role in an analysis, then the model may start writing as if it were an expert, instead of looking at the content soberly. Not ideal, unless you like expensive plays.

That's why prompting works better today if you start with the task at hand:

• Should AI write something?
• Should AI analyze something?
• Should AI analyze first and then write?

Only then do you decide whether a role, tone of voice or workflow is needed.

Does the language of your prompt matter?

Less than is often thought. English can be useful for technical terms, API settings or fixed labels such as Goal, Audience and Output. But if you want Dutch output, you can write your prompt in Dutch just fine.

More important than language is consistency. Don't mix Dutch and English unnecessarily. Do you use English labels? Fine. Then just fill in the content clearly in Dutch.

Example:

"Task: write a blog intro.
Goal: convince entrepreneurs to make their website faster
Target group: SME entrepreneurs with limited technical knowledge
Tone of voice: down-to-earth, practical and slightly cocky
Language: Dutch
Output: blog intro of up to 150 words"

No prompt poetry. Just clear instructions.

Our roadmap for the perfect prompt

The success of AI hinges on the quality of your prompts. The better you direct AI, the more relevant and useful the results will be. Our 10-step plan will help you create the most complete prompt possible.

1. Task

Start not with who AI should be, but with what AI should do. Should the model write something, analyze it, or analyze it first and then write it? By making the task sharp, you prevent AI from guessing for itself what the intent is.

A role or viewpoint can still help, but especially for writing tasks. For analysis, advice or factual questions, it often works better to describe desired behavior.

Example:

"Task: write a blog article about why website speed is important for SMBs."

Or for an analysis task:

"Task: analyze this landing page for SEO, conversion and clarity. Provide concrete areas for improvement."

 

2. Goal

Describe what you want to achieve with the output. Do you want to inform, persuade, inspire, activate or compare? A clear purpose helps AI make choices in structure, argumentation and CTA.

Example:

"Goal: Convince business owners that website speed is not just a technical issue, but directly affects user experience, findability and conversion."

 

3. Target Audience

Tell for whom the output is intended. Name function, knowledge level, need and any doubts or resistance. The more specific the target audience, the better AI can tailor the content and tone.

Example:

"Target audience: SME entrepreneurs and marketing managers with limited technical knowledge. They know their website is important, but still often see speed as something for developers."

 

4. Tone of voice

Specify how the text should sound. Words like professional, modern or fresh are often too vague. Make the tone concrete with descriptions or a short example sentence.

Example:

"Tone of voice: down-to-earth, practical and slightly opinionated. Write as if a savvy marketer is shaking up an entrepreneur without being pedantic."

 

5. Style

Highlight how the text should be written and structured. Consider language level, sentence length, use of jargon, headings, bullets, paragraphs and language. Preferably write your prompt in the same style as the output you want back.

Example:

"Style: write in Dutch at B1-B2 level. Use short paragraphs, clear H2 headings and bullets where useful. Use common Dutch words for general explanations. Use technical terms only when they are substantively necessary and then explain them briefly."

 

6. Output

Define what AI should produce. A blog, LinkedIn post, product page, analysis or e-mail all require a different structure. Also, name length and level of detail. This is a better place for length instructions than at the goal.

Example:

"Output: a blog article of about 900 words with a catchy intro, clear intermediate headings, practical examples, a brief explanation of Core Web Vitals, concrete tips and a concluding CTA."

 

7. Knowledge

Specify what information AI should use. Put behavior in the prompt, but prefer to put lengthy product information, brand documents, rates or customer cases in separate files. Also, name what AI should do when information is missing.

Example:

"Knowledge: use the supplied information on Core Web Vitals and MODX as the primary source. If information is missing, name it explicitly and don't make anything up."

 

8. Working Method

Describe a process only if the task is complex. For a long blog article, strategy or analysis, it may be smart to have AI create an outline first. For a short social post, that's usually overkill. Prompting should help, not become bureaucratic.

Example:

"Working method: create a short outline first. Then write the article. Then improve the text for clarity, flow and persuasiveness."

 

9. Requirements

Lay out hard requirements. Think SEO keywords, length, prohibited claims, legal requirements, trademark rules or technical prerequisites. Formulate positively whenever possible: say what AI should do, not just what it should not do.

Example:

"Conditions: use the keyword "website speed" naturally in the text. Write a maximum of 900 words. Avoid exaggerated claims such as "guaranteed more sales." Don't write a technical manual, but a practical article for entrepreneurs."

 

10. Validation

Have AI specifically check that the output is satisfactory. A general instruction like "check your answer" is too vague. Provide concrete criteria on which AI should check the output.

Example:

"Validation: check that the text matches the target audience, makes the importance of speed concrete, is technically correct, contains the main keyword and ends with a clear CTA."

 

This is how you end up with the following prompt:

Task
Write a blog article on why website speed is important for SMBs.

Objective
Convince business owners that website speed is not just a technical issue, but directly affects user experience, findability and conversion.

Target Audience
SME entrepreneurs and marketing managers with limited technical knowledge. They know their website is important, but still often see speed as something for developers.

Tone of voice
Sober, practical and slightly cocky. Write as if a smart marketer is shaking up an entrepreneur without being pedantic.

Style
Write in Dutch at B1-B2 level. Use short paragraphs, clear H2 headings and bullets where useful. Use common Dutch words for general explanations. Use technical terms only when they are substantively necessary and then explain them briefly.

Output
A blog article of about 900 words with a catchy intro, clear intermediate headings, practical examples, a brief explanation of Core Web Vitals, concrete tips and a concluding CTA.

Knowledge
Use the provided information about Core Web Vitals and MODX as the primary source. If information is missing, name it explicitly and don't make anything up.

Working method
First, create a short outline. Then write the article. Then improve the text for clarity, flow and persuasiveness.

Prerequisites
Use the keyword "website speed" naturally in the text. Write a maximum of 900 words. Avoid exaggerated claims such as "guaranteed more sales." Don't write a technical manual, but a practical article for entrepreneurs.

Validation
Check that the text connects with the target audience, makes the importance of speed concrete, is technically correct, contains the main keyword and ends with a clear CTA.

 

This is how you make your prompt specific, give the AI context and keep the output consistent.

Examples of good and bad prompts.

Each department within your company can use AI tools like ChatGPT and Claude differently. The basics remain the same: The sharper your command, the better the output. That's why we make the roadmap concrete with examples for marketing, sales and HR.

Marketing

❌ Bad prompt: "Write a landing page for our Google Ads campaign about our marketing tool."

Why doesn't this prompt work?

• The task is too vague: it is not clear what exactly the landing page is supposed to do.
• The purpose is missing: should the page inform, persuade or generate leads?
• The target audience is missing: AI does not know who is reading the page or what problem they are experiencing.
• The structure is missing: AI has to decide how the page will be built.
• The terms are missing: there is nothing about SEO, CTAs, claims or desired length.

✅ Good prompt:
"Task.
Write a Dutch landing page for a Google Ads campaign about our marketing tool.

Objective
Convince SME entrepreneurs and marketers to try our marketing tool for free. Show how the tool helps make Google Ads campaigns more efficient and ad budget smarter.

Target Audience
Small to medium-sized businesses that use Google Ads but struggle with high costs, low conversion rates and little insight into what their campaigns are delivering.

Tone of voice
Motivational, professional and direct. Write practically and persuasively, without exaggerating.

Style
Write in Dutch at B1-B2 level. Use short paragraphs, clear H2 headings and bullet points where useful. Use common Dutch words for general explanations. Use marketing terms only when substantively necessary and explain them briefly.

Output
Create a landing page of about 400 words with:
- a powerful H1 that identifies problem and solution
- a short intro of 20 to 50 words about inefficient ads
- three benefits of the tool in bullet points
- a short customer case with concrete results
- a clear CTA to try the tool for free

Knowledge
Use the information provided about the marketing tool as the primary source. If information is missing, name it explicitly and do not make up features or results.

Working method
First, create a short outline of the page. Then write the landing page.

Prerequisites
Use the keywords "Google Ads optimization," "increase ad ROI" and "marketing tool for small businesses" naturally in the text. Always link claims about results with substantiation or formulate them carefully.

Validation
Verify that all three SEO keywords are naturally incorporated, the text matches the target audience and the CTA is clear and specific."

 

Sales

❌ Bad prompt: "Write a presentation about our product and why it's useful."

Why doesn't this prompt work?

• The task is unclear: Is it about a sales deck, pitch text or slide structure?
• The goal is missing: AI doesn't know what action the reader or listener should take.
• The target audience is not specific: a CFO, IT manager and logistics manager have different questions.
• Benefits are not linked to pain points or results.
• The desired structure and CTA are missing.

✅ Good prompt:
"Task.
Write a Dutch sales presentation on transportation management software (TMS).

Objective
Convince logistics managers that the software helps them work more efficiently, manage transport processes better and reduce costs. The presentation should lead to a demo request.

Target group
Logistics managers in medium-sized companies. They know the operational challenges, such as delays, error-prone planning and high transportation costs, but have limited technical knowledge.

Tone of voice.
Professional, solution-oriented and persuasive. Write as if you are thinking along with the logistics manager, not as if you are reading up on a product sheet.

Style
Write concisely in Dutch at B1-B2 level. Use clear headings, short sentences and bullet points. Use technical terms only when necessary and explain them briefly.

Output
Create a sales presentation of about 400 words with:
- a powerful opening sentence that immediately evokes recognition
- a brief introduction of the problem
- three to five advantages of the software in bullet points
- a short practical example with concrete results
- a clear CTA to request a demo

Knowledge
Use the provided product information as the primary source. If customer results or figures are missing, do not use made-up percentages or cases.

Working method
First, create a brief structure for the presentation. Then work out the text for each section.

Prerequisites
Keep the presentation compact and convincing. Always link benefits to recognizable logistical challenges. Make result claims only if they are substantiated.

Validation
Check that the presentation is relevant to logistics managers, the benefits are concrete and the presentation ends with a clear demo CTA."

 

HR

❌ Bad prompt: "Write a job posting for a software developer."

Why doesn't this prompt work?

• The position is too general: junior, medior and senior developers are looking for something different.
• The purpose is missing: should the text inform, enthuse or generate direct applications?
• The target audience is not fleshed out: AI does not know what is important to this candidate.
• The tone of voice is missing: as a result, the text can quickly become generic.
• The terms are missing: there is nothing about SEO, content, length or CTA.

✅ Good prompt:
"Task.
Write a Dutch job posting for a senior software developer.

Objective
Excite experienced software developers to apply by showing that the position offers technical challenge, autonomy and influence on projects.

Target Audience
Senior software developers with at least five years of experience. They are looking for a role where they can leverage their expertise, get to make technical choices and work on challenging projects with a healthy work ethic.

Tone of voice.
Professional, motivational and direct. Address the candidate personally and show that their experience is taken seriously. Don't make the text too formal.

Style
Write in Dutch at B1-B2 level. Use short paragraphs, clear headings and bullet points for tasks and benefits. Use technical terms only where they are relevant to the job.

Output
Create a job posting of about 400 words with:
- a catchy opening sentence that arouses curiosity
- a brief description of the role and why it is of interest to senior developers
- three to five key responsibilities
- three to five benefits of working at the company
- A clear CTA to apply

Knowledge
Use provided job information and terms of employment as a source. If information on salary, stack, team or working method is missing, name it as a point of interest and do not fill it in yourself.

Procedure
First make a short outline. Then write the job description.

Requirements
Use the keywords "senior software developer job", "experienced developer wanted" and "software development job" naturally in the text. Use concrete benefits instead of vague promises such as "competitive salary" or "dynamic work environment."

Validation
Check that the text addresses senior developers directly, contains all three SEO keywords naturally and ends with a clear job application CTA."

 

Tip!
Let AI critically review your job posting text from the needs of your ideal candidate. First, describe who that candidate is and what they look for. Then let AI name the biggest drop-off points and rewrite the job posting.

 

Our tips for getting started yourself.

Whatever department you work in, an AI tool is only as good as the instructions you give it. Think of it as a new colleague you're inducting: the clearer you explain what you expect, the more useful the result.

Whether you use our roadmap or not, these five rules will help you get better output.

1. Start with the task: First, make clear what AI is supposed to do. Write, analyze or both. So don't automatically start with a role, but with the task.
2. Provide context: Explain why you need the output, who it is for, and what background information is important. AI can do a lot, but mind reading is not yet on the roadmap.
3. Be specific: Vague prompts produce vague output. Name target audience, tone of voice, structure, length, sources and terms. The more specific you are, the less AI has to guess.
4. Keep instructions consistent: Don't give conflicting commands. So don't ask for "short and sweet" and "very comprehensive with lots of examples" at the same time. Pick what is important and put hard requirements clearly.
5. Test and improve: A prompt is rarely perfect overnight. Test your prompt, look at the output and adjust one thing at a time. Feel free to ask AI to improve your prompt, but provide criteria.

With this approach, you'll get more out of AI and avoid wasting time on disjointed, generic content. Want to get started yourself? Download our prompting cheatsheet and build prompts that AI does understand. No magic. Just better instructions.