Using AI: Prompt Competence

On this page, you’ll learn how to write prompts for generative AI that will make the output more useful, accurate and verifiable. This information is intended for students, teaching staff, researchers, and members of administrative staff.

The so-called prompt is what is entered into generative AI in order to achieve a specific result. The quality of the prompt is crucial to the AI’s output. A simple prompt will usually produce superficial results. So-called mega-prompts can be used to achieve better results in the productive use of generative AI. Here, the generative AI is e.g. assigned a role, and given context and a task. Mega-prompts may also contain information regarding the desired structure of the output and intermediate steps. The use of delimeters (e.g. #, ##, *, _) makes it easier for an AI to understand a prompt. For example, you might enter: “#Role: ‘You’re a lecturer at a German university.’

You can create your own mega-prompt or use prompt templates. Prompt templates are pre-formulated text prompts that you can customise to suit your needs. They help you formulate clear instructions and, as a result, achieve better, more precise output. Prompt templates provide valuable guidance and make it easier to get started, particularly if you have little experience in using AI or need a result for specific circumstances.

Components of the mega-prompt

• #Role: Who or what should the AI tool simulate?
• #Task: What needs doing?
• #Steps: What are the individual steps, and in which order should they be carried out?
• #Context or restrictions: What needs to be kept in mind here?
• #Objective: What specific objective is the dialogue intended to achieve?
• #Output format: How should the bot’s response be presented?

The Mega-Prompt breaks a query down into six parts. #Role specifies the role in which the AI should respond; for example, as a tutor or as a critical reader of a paper. #Task describes exactly what is required, for example create an outline, review a paragraph or compare arguments. #Steps help the AI to proceed in a structured manner if the overall task consists of several individual steps. #Context provides the necessary background information. #Objective clarifies the expected output. #Format determines how the response should be presented, e.g. a tabular overview, etc.

Components of an ACDQ formula

• Act: Act like... [e.g. an employee in the xy department]; in which role should the AI act?
• Context: What contextual information does the AI need to take into account?
• Deeply: Ask the AI to ‘think deeply about the answer’.
• Questions: Ask the AI to ask any follow-up questions – what does the AI need to know in order to provide an accurate answer?

General information

AI tools can make many aspects of studying easier – from exam preparation and structuring papers to understanding difficult academic texts. However, the following applies: never enter personal data, marks, the names of fellow students, or internal course materials provided by lecturers into external AI systems. Texts may only be uploaded if they are freely accessible (Open Access or Creative Commons). Typing or uploading copyright-protected text and images may have legal consequences. Please also take note of the basic training on the EU AI Act [de] and the university guidelines [de] on the use of AI.

AI is no substitute for independent thinking

AI systems generate texts that sound convincing, but are not necessarily correct. The phenomenon is called a hallucination: AI invents facts, figures or sources that do not actually exist, without hesitation or indication of any uncertainty. You must independently verify every source reference, every subject-specific term, and every statement that you wish to use from an AI output. AI can assist with structuring, formulating and analysing text, but it does not replace independent reading, comprehension and the critical evaluation of the content. Anyone who skips these steps not only risks making factual errors and committing plagiarism, but also risks their own understanding being built on sand.

Better prompts

The quality of the AI output heavily depends on how precisely the query has been formulated. If you simply ask the AI, “Explain to me…”, you’ll get a generic summary. On the other hand, if you explain that you’re studying philosophy in your third semester, are preparing for an exam on the basic concepts of epistemology, and are looking for a clear explanation in understandable terms, you’ll get a much more useful answer. To achieve good results, use the Mega-Prompt or the ACDQ formula.

General information

• Please take note of the Basic training on the EU AI Act [de]

Better prompts

The quality of the AI output heavily depends on how precisely the query has been formulated. If you simply ask the AI to “Give me exercises for students…”, you will receive a generic list of various exercise formats that are not tailored to the specific context. On the other hand, if you indicate that you require methods to engage students in an online seminar, you will receive a much more useful answer. Use the Mega-Prompt, the ACDQ formula or the prompt templates provided here for good results.

General information

• Please take note of the Basic training on the EU AI Act [de]

Better prompts

The quality of the AI output heavily depends on how precisely the query has been formulated. If you simply ask the AI to “Give me ideas for my research…”, you will usually receive a general list of possible suggestions that are not relevant to the current state of research, the methodology, the discipline, or the specific research question. The output becomes much more useful if you provide the context clearly, for example: ‘I am planning a qualitative interview study on students dropping out during their first year of studies.’ ‘Develop five possible research questions that can be investigated empirically using guided interviews, and provide a brief justification for each question.’ If you formulate your queries this way, the AI can provide a much more precise, research-relevant answer. Use the Mega-Prompt or the ACDQ formula for good results.

Use case: qualitative social research – support for drafting an interview manual

The background for the creation of such a manual is that interviewees always interpret questions according to their own experiences, personal backgrounds, or keywords that are relevant to them. Consequently, even among people with similar professional or personal backgrounds, questions can be interpreted in very different ways. AI can help here by better anticipating such potential misunderstandings for different groups of people and refining the questions accordingly. This is also important for breaking out of one’s own academic tunnel vision, because understandings of terms, implicit assumptions, and things taken for granted within the discipline can easily creep into the way questions are phrased without being noticed. Correspondingly, the AI is therefore instructed to adopt the perspectives of several individuals from the group being surveyed, each with different backgrounds, and to examine how a specific interview question might be understood differently, ambiguously or even incorrectly. 

General information

• Please take note of the Basic training on the EU AI Act [de]
• Please take note of the University Guidelines on the Use of AI [de]
• As a general rule, do not enter any sensitive data or information into AI systems hosted outside the university.
• Text may only be entered into Large Language Models (LLMs) where this is expressly permitted (open access and creative commons). Uploading copyright-protected text and images to an LLM may have legal consequences.
• In order to generate role and context-specific AI output, we recommend using prompt templates that provide precise details regarding the context and the desired output. Use Mega-Prompts or an ACDQ formula for such purposes.

Tips for using the prompt templates

The following prompts assume that you have provided the AI with specific contextual and role-related information related to your needs. The prompts set out here are intended as inspiration for the wide variety of tasks in public administration. The sections where you could insert your own text, web links and the like are marked in [square brackets].