You get the best results from ChatGPT and other AI tools when you give precise, structured instructions and align your request clearly with a concrete goal. Prompting is essentially the art of telling an AI exactly what to do, in what context, and in what format. The following sections cover the most important techniques, from basic approaches to advanced methods for professional use.<\/p>\n\n\n\n
A prompt delivers good results when it has a clear objective, provides sufficient context, and contains no contradictory or vague instructions. The mental groundwork done before actually writing the prompt is the decisive lever \u2014 more powerful than any single technique.<\/p>\n\n\n\n
Before you type a prompt, you should ask yourself three questions: What exactly should the result achieve? Who is it intended for? And in what format should it appear? If you clarify these questions in advance, you will automatically write better instructions.<\/p>\n\n\n\n
Common mistakes that reduce the quality of responses include:<\/p>\n\n\n\n
If you avoid these pitfalls and approach the prompt with a concrete picture of the desired result in mind, you will receive significantly better responses \u2014 regardless of which AI tool you use.<\/p>\n\n\n\n
Zero-Shot Prompting means you give the AI a task without providing an example. One-Shot provides a single example, and Few-Shot provides several. The more relevant examples you include, the more precisely the AI can imitate the desired format, style, or structure.<\/p>\n\n\n\n
In practice, it looks like this: A Zero-Shot prompt would be „Write a product description for a project management tool.“ A Few-Shot prompt would include two or three finished example descriptions so the AI can adopt the style, length, and tone.<\/p>\n\n\n\n
Few-Shot Prompting is especially useful when you need consistent outputs \u2014 for example, for recurring reports, marketing texts in a specific style, or standardised document templates. For simple, one-off requests, Zero-Shot is often perfectly sufficient. The extra effort of providing examples pays off most with complex or format-bound tasks.<\/p>\n\n\n\n
Chain-of-Thought Prompting means explicitly asking the AI to explain its approach step by step before arriving at an answer. This approach significantly improves quality for complex, analytical, or multi-step tasks, because the model makes fewer reasoning errors when it explicitly works through its solution path.<\/p>\n\n\n\n
A simple example: instead of „What risks does this project have?“, you ask „Analyse the following project data step by step and derive the three greatest risks from it.“ The AI structures its thinking before making a judgement and delivers more transparent results.<\/p>\n\n\n\n
Chain-of-Thought is particularly worthwhile for:<\/p>\n\n\n\n
For short, straightforward requests such as rephrasing a sentence, this technique is not necessary and can make the response unnecessarily long.<\/p>\n\n\n\n
When you assign the AI a role \u2014 for example „You are an experienced marketing strategist“ \u2014 this noticeably influences the perspective, style, and depth of the response. The AI uses the role profile as a filter to generate more relevant, domain-specific outputs.<\/p>\n\n\n\n
This principle lies at the heart of the RACE technique, a proven structural framework for good prompts. RACE stands for Role, Action, Context, and Example. A complete RACE prompt might read: „You are a world-leading expert in process automation. Write a short guide for executives who want to introduce AI tools into their daily work. Our company has 50 employees and does not yet use any AI-supported workflows. Orient yourself to the style of the attached example document.“<\/p>\n\n\n\n
Role assignment is particularly effective when you need expertise from a specific domain or when the response should be tailored to a particular target audience. Combined with the right context and a clear goal, it is one of the simplest ways to significantly improve output quality. Those who attend structured AI training courses<\/a> learn such techniques using concrete application examples from everyday professional life.<\/p>\n\n\n\n When a prompt does not deliver the desired result, the most effective approach is to analyse the response and then deliberately change a single parameter: add more context, refine the role, explicitly specify the format, or break the task down into smaller steps.<\/p>\n\n\n\n Iterative prompting works like a feedback conversation. After an unsatisfactory response, you can tell the AI directly what was missing: „The answer was too general. Focus on medium-sized companies with fewer than 200 employees and name three concrete measures.“ This follow-up prompt is often more effective than starting over with a completely new attempt.<\/p>\n\n\n\n A particularly clever method is Meta-Prompting: here you ask the AI what it still needs in order to solve the task optimally. Example prompts for this would be:<\/p>\n\n\n\n Anyone who regularly tests, refines, and documents prompts builds up a personal library of working templates over time \u2014 which significantly reduces the effort involved in recurring tasks.<\/p>\n\n\n\n In a professional context, structured techniques such as RACE, Few-Shot Prompting, and the system prompt are especially valuable, because they deliver consistent, reusable results and can be integrated into existing workflows.<\/p>\n\n\n\n The system prompt \u2014 also described as an onboarding briefing for the AI \u2014 is a one-time pre-configuration of the AI model. Here you define how the AI should communicate in general: level of detail, directness, format, target audience. All subsequent requests are then answered within this framework, without you having to re-explain your preferences each time. This principle underlies Custom GPTs or priming in other AI platforms.<\/p>\n\n\n\n For typical professional use cases, the following techniques are recommended:<\/p>\n\n\n\n The choice of the right model is also crucial. For simple text tasks, faster, smaller models are sufficient. For complex analyses or multi-step tasks, it is worth using more powerful reasoning models that work more slowly but with significantly greater precision. Anyone who wants to build AI competence in a structured way<\/a> should think about both dimensions \u2014 prompting technique and model selection \u2014 together.<\/p>\n\n\n\n We at mITSM offer practical training that starts exactly where professionals and managers encounter AI in their daily work. Our trainers are certified experts who actively use AI tools themselves and bring their experience directly into the training.<\/p>\n\n\n\n Our offering in the area of AI competence includes:<\/p>\n\n\n\nHow can you iteratively improve prompts when the results are not right?<\/h2>\n\n\n\n
\n
Which prompt techniques are particularly suited to professional use?<\/h2>\n\n\n\n
\n
How mITSM supports the development of AI competence<\/h2>\n\n\n\n