Abstract:

While generative AI is rapidly entering higher education, many institutions struggle to ensure reliable, consistent, and pedagogically meaningful AI responses for staff and students, especially for those who lack prompt-engineering expertise. This paper introduces a curated prompt library as a governance mechanism that allows universities to standardize and guide AI interactions while keeping conversational AI systems easy to use. The basic library is implemented as a structured, tabular collection of predefined prompts designed and maintained by the institution to address recurring academic and administrative tasks. Unlike static prompt collections, the library is directly integrated into the campus management system, allowing prompts to be selected and executed interactively. By providing prepared and directly usable prompts, institutions ensure that interactions with large language models follow consistent structures, institutional guidelines, and pedagogical intentions while significantly lowering the entry barrier for staff and students. Furthermore, the system supports institutional governance by allowing administrators to control which user groups can access specific prompts and to monitor their usage across the platform.

The library distinguishes five prompt types:
- Firstly, open text prompts allow users to freely modify the prompt and support exploratory use of AI, for example when creating a social media post with pictures.
- Secondly, restricted prompts limit modifications to predefined parameters while keeping the core prompt structure fixed, ensuring consistent results. Such prompts are used, for example, for generating structured teaching materials such as case studies where only specific elements (e.g., topic) may be adjusted.
- Thirdly, database-based prompts dynamically incorporate structured institutional data, which allows the AI to generate responses that refer to contextual information such as the structure of a course according to the module handbook.
- Fourthly, document-based prompts use institutional documents, such as official learning materials, in order to support context-aware responses and enable AI-supported learning dialogues.
- Finally, agent prompts perform more complex tasks and external communication. Based on the Model Context Protocol (MCP), for example, an agent can conduct literature searches by communicating with external academic databases, compare relevant publications, and propose curated reading lists for further study.

The prompt library is also integrated into a campus chat assistant within the institutional environment. During natural language interactions, the assistant analyzes the user’s intent and retrieves suitable prompts from the library. For example, when a lecturer asks for support in preparing a teaching session, the assistant looks up the lecturers appointments, identifies the study group and topic, selects relevant prompts and enriches them with contextual information such as the module descriptions from the module handbook. The assistant not only presents appropriate prompts as suggestions but is also able to execute them directly, enabling users to benefit from curated prompt logic and institutional knowledge without needing to formulate complex prompts themselves.

By combining curated prompts, contextual campus data, and agent-based workflows, the presented framework demonstrates a practical and scalable approach to reliable and governed AI assistance in higher education.

Keywords:

Prompt Governance, Curated Prompt Libraries, Generative AI in Higher Education, AI, MCP, RAG.