A new research synthesis by the Federal Ministry of Education debunks myths and shows that artificial intelligence (AI) can help schools move forward, but only under clear conditions.

Since ChatGPT became publicly available in November 2022, the debate surrounding artificial intelligence in German schools has fundamentally shifted. What initially sparked panic—students having AI do their homework—has evolved into a more nuanced discussion. The latest research synthesis, “Artificial Intelligence in Schools,” by Katharina Scheiter and her team provides the first comprehensive overview of the current state of affairs in both academia and practice. The conclusion: AI is neither a silver bullet nor the downfall of the education system. Rather, it is a tool that requires targeted use and expert guidance.

According to a representative study by the Vodafone Foundation, 73 percent of the students surveyed believe that AI offers more opportunities. At the same time, 38 percent reported that AI plays no role at their school. This discrepancy between potential and reality characterizes the German school system. While other countries are systematically introducing AI tools, Germany often lacks the necessary strategies, equipment, and professional development opportunities for teachers.

Data protection is not an obstacle, but a challenge to be addressed

The most common objection to AI in the classroom is data privacy. In fact, many free AI applications, such as ChatGPT, are problematic in their free versions because they store user data on non-European servers and use it to train their models. However, the research synthesis makes it clear: “Neither the GDPR nor, in the future, the AI Act fundamentally prevent the use of AI in schools; rather, they specify which applications may be used and under what circumstances.”

Several federal states have now responded and are offering solutions that comply with data protection regulations. Baden-Württemberg provides its own interface, fAIrChat, while Rhineland-Palatinate and Mecklenburg-Western Pomerania have purchased statewide licenses for providers such as “Fobizz” or “schulKI.” These allow students to access AI tools without personal data being shared. Education-specific applications such as the writing assistant “FelloFish” or the intelligent tutoring system “bettermarks” were also developed to be GDPR-compliant from the outset. The researchers’ message is clear: it is not data protection that prevents the use of AI, but a lack of expertise and inadequate infrastructure.

Teachers remain indispensable

A key concern among many educators: Will AI make teachers obsolete? The clear answer from the scientific community is: No. “Even fully automated tutoring systems do not work on their own,” the study emphasizes. The pedagogical and didactic integration of AI technology remains crucial. Teachers must embed AI tools into an overall concept combining analog and digital learning methods, evaluate the quality of the results, and make corrections as needed.

Meta-analyses show that intelligent tutoring systems (ITS) achieve learning outcomes comparable to those of small-group instruction or individual tutoring. They are particularly effective for learning new material. However, it is crucial that AI complements traditional instruction rather than replacing it. Based on meta-analytic findings, the research synthesis concludes that ITS are more effective when used in combination with traditional instruction than when they replace it. Thus, AI does not render teachers redundant; rather, it allows them to focus more on building relationships, providing individualized support, and designing challenging learning scenarios.

AI as a tool for reducing workload

Another finding: AI can significantly reduce the workload for teachers, provided it is used correctly. The recently published essay by Heine et al. (2024) lists 17 areas of responsibility, ranging from lesson planning and creating materials to grading assignments. Generative AI like ChatGPT can, for example, create texts at various difficulty levels, generate worksheets, or translate letters to parents into multiple languages. With its AI Toolbox, Cornelsen Verlag already offers a GDPR-compliant tool that accesses verified publishing content and supports teachers with pre-written prompts.

However, the researchers caution against unrealistic expectations. The study by Heine et al. shows that many teachers consider AI-generated solutions to be not immediately usable. The problem is that people often expect an initial prompt to deliver perfect results right away. In reality, working with AI is more like a collaborative process in which prompts need to be refined step by step. Those who view AI as a co-author rather than an autopilot can better leverage its strengths and save time despite the revision effort.

Assessing Learning Outcomes in the Age of AI

Perhaps the greatest challenge lies in the area of performance assessment. Studies show that OpenAI’s “GPT-4” passes both parts of the U.S. bar exam and scores in the 93rd percentile on the SAT. In psychology exams at British universities, 94 percent of AI-generated exams were not identified as such and scored, on average, half a grade higher than those of human students. The question, then, is no longer whether AI can handle exams, but how educational institutions will respond to this.

Teachers Joscha Falck and Hendrik Haverkamp have developed a five-dimensional model for this purpose: exams can be conducted “about, with, using, despite, and without AI.” While exams that do not allow the use of aids remain valuable, skills in working with AI should increasingly be assessed as well. For example, by having students critically analyze AI-generated texts, optimize them, or use them as a starting point for their own work. Accordingly, the Standing Conference of the Ministers of Education and Cultural Affairs recommends new exam formats, such as the presentation and defense of projects created in collaboration with AI. This reflects the reality of modern working life—where AI has long been part of everyday life.

What teachers need to learn now

The good news is that teachers do not need programming skills or a deep technical understanding of neural networks to make effective use of artificial intelligence. The research synthesis identifies six key areas of competence, ranging from basic theoretical knowledge to ethical reflection and ongoing professional development. What is crucial is an understanding of how AI works, where its limitations lie, and what biases may be present in the training data.

After all, AI is no more objective or error-free than humans. On the contrary: algorithms adopt the biases of their training data. A real-world example: When asked to name the ten most important philosophers, ChatGPT generates a list that includes not a single woman and consists exclusively of European thinkers. Teachers must be able to recognize such biases and discuss them with their students. The Standing Conference of the Ministers of Education and Cultural Affairs therefore recommends embedding AI competencies in all three phases of teacher education: from university training through the teaching internship to continuing education and professional development.

The research synthesis “Artificial Intelligence in Schools” (Scheiter et al., 2025) shows that AI can enrich instruction and reduce the workload for teachers, but only when used thoughtfully. The biggest hurdles are not technical, but rather organizational and pedagogical in nature. What is needed now is widespread access to equipment, systematic professional development, and clear strategies so that AI can realize its full potential.

About the study: The research synthesis “Artificial Intelligence in Schools” was published in May 2025 by the Federal Ministry of Education, Family Affairs, Senior Citizens, Women, and Youth as part of the AI support process within the framework program for empirical educational research. The team of authors led by Katharina Scheiter combined scientific evidence with practical school experience, involving experts from the field in workshops. The study addresses key questions regarding the functioning, availability, learning effectiveness, and potential applications of AI in a school context and is explicitly intended as a snapshot in a highly dynamic field of research.

Cover image: © Pcess609