Clever prompting: how it works! Tips for optimized automation with AI

A good prompt is crucial for business

An AI can only provide accurate information down to the last detail and do exactly what the company wants and needs if it has a good prompt. This is the only way to ensure that customers are addressed correctly and consistently. This also means that the AI must be able to make small decisions, such as which department a request should be forwarded to. The AI must also be able to reject requests that do not fit the desired context. Here are a few tips to make sure everything works.

1. Make only one specific request per prompt

A good prompt should contain only one question or request, and it must be as specific as possible. To do this, you can use the COSTAR framework: This means specifying exactly what the AI should deliver in six steps. First, define the context, i.e., the role and situation, for example: You are a scientist or customer. This is followed by the definition of clear and unambiguous objectives and, if necessary, what the AI should not do. Next, the individual steps are outlined so that the AI can break down the request into individual solution steps and solve it step by step. After that, the tone and style should be defined, as well as the audience, i.e., the target group. Finally, the format of the response is specified.

2. Prompting works best step by step

Prompting is an iterative, creative process. This means that users should critically examine the responses and evaluate what worked and what didn't. The next step is to improve and deepen the response with a new or expanded prompt. The aim is to systematically try out inputs in a live setting and modify them again and again to obtain the optimal response. This allows you to test step by step which prompt works best for the specific use case. Step by step, you can move from a simple prompt to an optimized query.

3. Avoid excessive wording in prompt design

It is crucial to understand how to formulate inputs efficiently in order to achieve your goal quickly and precisely. However, there is no universal recipe, because every task is different. That's why experimentation is always part of prompting. Nevertheless, the following applies in every case: the quality of the input determines the quality of the result. When prompting, you should therefore be precise and targeted and avoid excessive wording or vague requests. Ultimately, targeted prompt design can achieve reliable, repeatable results.

4. Simply ask the AI and let it prompt

The AI itself can assist with prompting by asking it directly for better formulations. For example: “Give me five different prompts for the correct solution to a service request.” In this way, the AI provides detailed variations and creative input that helps to create more precise instructions. Use the “step back prompting” technique for this. As in the first point, give specific instructions to create a prompt, which you can then use again. Because AI models have been trained with millions of texts, they know what kind of questions lead to better results. Having prompts prompt you can also save time.

5. Make requests in English

Since many AI models, such as ChatGPT, have been trained primarily with English-language data, entering prompts in English can lead to more precise and better results. The text generated in English can then be translated into German using AI translators.