How to Prompt AI in Multidisciplinary School Teams

Artificial intelligence (AI) is increasingly finding its way into everyday school life and offers enormous potential to support teachers, childcare educators, social workers, and other educational professionals. At the heart of effective AI use – for example with large language models (LLMs) such as ChatGPT – is "prompting": the ability to formulate clear, goal-oriented input requests. A well-designed prompt can make the difference between whether an AI provides valuable support for documentation, lesson preparation, or the creation of support plans, or merely delivers generic and unhelpful responses.

This article serves as a practical guide to developing prompting skills specifically for the school context. It explains what prompting means and why it is an important competency for multidisciplinary teams in schools. The recommendations presented here are based on current insights from leading AI developers. The goal is to convey practical knowledge so that AI models can be used confidently and result-oriented in a demanding professional environment.

Understanding the basics: how AI interprets requests

To make the most of AI's potential, a basic understanding of how it works is helpful. LLMs are highly developed systems that have learned to recognize patterns from vast amounts of text. Simplified somewhat, they generate responses by predicting, based on the input, the most likely sequence of words.

An important distinction: general AI chatbots do not have access to specific internal data from a school or institution. Some of the examples given here – especially those that refer to individual student information – assume that an AI system is being used that has been specifically developed for the educational sector and can securely access a protected internal knowledge store (e.g., digitized notes or documentation within the school). Such systems, often referred to as knowledge-based AI systems or augmented language models, frequently use advanced technologies to improve their information generation capabilities. One of these key technologies is Retrieval-Augmented Generation (RAG). RAG systems are designed to base their responses not only on internally trained knowledge, but also to retrieve specific, context-relevant information in real time from knowledge sources.

This typically happens in several steps: first, the system analyzes the user query to identify relevant keywords or concepts. A retrieval module is then activated to search large databases, document collections, or the internet for relevant information. This retrieved information is then made available to the generation module (often a large language model). The generation module uses this additional information to formulate a more precise, better-grounded, and contextually more relevant response than it would be able to produce from its pre-trained knowledge alone.

The advantage of RAG lies in its ability to reduce hallucinations and significantly increase the currency and accuracy of generated content. Rather than delivering general or potentially outdated information, these systems can point to specific, verifiable facts. This is particularly useful in areas where precision and reliability are critical – such as medical diagnostics, legal advice, or education.

Another important concept in all LLMs is the context window: think of it as the AI's short-term memory. It has a limited capacity, which means that in the case of very long requests or extended conversations, earlier information can be lost. For this reason, concise and well-structured prompts are often more effective.

The challenge of prompting lies in formulating the often complex pedagogical intent in such a way that the AI can implement it precisely, given that AI takes language literally and does not have human everyday understanding or the ability to "read between the lines" – unless explicitly instructed to do so.

Effective Prompting Principles for School Teams

A few fundamental principles are crucial for the successful use of AI systems:

A. Clarity and specificity:

Requests should be formulated as clearly and precisely as possible.

• Vague (with a general AI): "Please give me general information on the topic of learning difficulties."
• Specific (with a general AI): "Explain the most common signs of dyscalculia in primary school children and name three pedagogical approaches to support them."
• Specific (with AI that has access to internal notes): "Summarize the observations about Max Mustermann from the past 3 weeks, focusing on notes that describe learning difficulties."

B. Providing sufficient context:

Giving the AI relevant background information helps it understand the request in the right framework.

• Without context: "Create a support plan."
• With context (with a general AI): "Draft a structure for a support plan for a Year 3 student with a diagnosed reading and writing difficulty. The plan should include sections for current status, goals, measures, and evaluation."
• With context (with AI that has access to internal notes): "Create a draft support plan for student Mira Muster, based on her documented learning progress in German and the goals from the last support plan meeting."

C. Defining the desired outcome: format, tone, length, role

Telling the AI what the output should look like is crucial.

• Without definition: "Write something about the trip."
• With definition: "Write a short report (approx. 150 words) on the class trip to the natural history museum for the school website. The tone should be positive and informative. The target audience is parents."

D. The power of instructions: guiding the AI step by step

For more complex tasks, such as preparing a parent meeting, guiding the AI through sub-steps is helpful.

• Unstructured: "Help me prepare a parent meeting."
• Structured (with a general AI): "Draft a guide for a parent meeting with the following sections: 1. Positive feedback on the child's behavior. 2. Presentation of learning areas that require attention. 3. Joint agreement on support measures."
• Structured (with AI that has access to internal notes): "Prepare the parent meeting for Lisa Meyer. 1. List positive developments in her work behavior over the past 6 weeks. 2. Name two current challenges in mathematics. 3. Formulate two concrete suggestions for support at home."

These principles are universal and are recommended by most AI developers.

Table 1: Basic principles of effective prompting in the school context

(Note: Some examples assume AI systems that can access specific internal school data.)

Data Protection When Using AI at School

Using AI in the school environment requires particular care when personal data is involved. Before entering information about students, parents, or colleagues, a school must assess the specific service, purpose, data categories, legal basis, authorization, configuration, and applicable contractual terms. A tool's provider or product category alone does not determine whether a particular use is permissible.

When selecting an AI tool, review its data processing agreement where applicable, processing locations, subprocessors, access controls, retention rules, and whether customer content may be used for model training. Specialized school products may provide controls for these needs, but the controls and their scope must be verified for the exact product tier and deployment. When in doubt, consult the school or school authority's data protection officer.

Insights from the pioneers: what AI developers advise

Developers of AI models such as OpenAI (ChatGPT), Google AI (Gemini), and Anthropic (Claude) give similar advice: clarity, context, definition of the desired output, and iterative refinement are decisive. OpenAI often emphasizes "showing through examples" and instructing the model to "think step by step." Anthropic highlights the value of assigning a role ("role prompting"). These overarching recommendations are also relevant for using specialized AI assistants in education – always with due regard for the data protection principles mentioned above.

Advanced Strategies for Complex School Tasks

With a little practice, more complex requests can also be addressed using AI:

A. Role-playing (persona prompting)

Assigning a role to the AI can adapt its tone and content.

• Example: "You are an experienced special education teacher. Give me feedback on this draft of a support goal for a student with autism spectrum disorder, based on general pedagogical principles."

B. Chain-of-Thought (CoT) / step-by-step reasoning

The AI is asked to make its thinking process transparent.

• Example: "A student has difficulty solving word problems in mathematics. Explain step by step what general diagnostic considerations and support approaches might be appropriate."

C. Iterative refinement: the art of improvement

The first prompt is rarely perfect. Prompts should be tested, analyzed, and adjusted until the result meets expectations. This is particularly useful when drafting general text modules for report card phrasing or parent letters that can then be individually adapted.

Table 2: Comparison of advanced prompting techniques for schools

Common Pitfalls to Avoid

The following typical sources of error should be kept in mind:

• Vagueness and ambiguity: "Help me with difficult students." More precise: "List three general de-escalation strategies for conflicts in the classroom."
• Overly complex prompts: too many instructions at once can overwhelm the AI. Better to break them into sub-steps.
• Leading questions or unconscious bias: "Write positive feedback for a student who often disrupts but sometimes has good moments." More neutral: "Describe how to formulate differentiated feedback that takes into account both a student's strengths and areas for development."
• Assuming background knowledge (with general AIs): the AI does not know school-specific abbreviations or particular local circumstances.
• Ignoring the iterative process: it is worth persisting beyond the first attempt and treating prompting as a process of experimentation.
• Not specifying the format: if a list or table is needed, this should be explicitly requested (e.g., "as bullet points," "as a table with the columns 'Strategy' and 'Target group'").

Table 3: Common prompting pitfalls in the school context and their solutions

Conclusion: Responsible AI Use in School Teams

The ability to craft effective prompts is an important key to unlocking AI's potential for the demanding daily work of school life. This guide has presented principles, strategies, and approaches for using AI systems as helpful assistants for documentation, planning, communication, and individual support – always with a commitment to responsible and GDPR-aligned use.

It is important to bear in mind: prompting is a learning process. With each application and adjustment of requests, confidence and the quality of results grow. Human expertise – pedagogical knowledge, the ability to define goals, and the capacity to critically evaluate results – remains indispensable. Effective prompting is the interface at which human intelligence and experience guide artificial intelligence, to best support professionals in multidisciplinary teams and to create space for direct educational work with children and young people.