Prompting
Prompting means giving clear instructions to an AI so it can deliver useful responses. The more precisely you describe the task, the context, and your expectations, the better the result will be. Think of the AI as a skilled colleague who does not already know your work.
Below, we introduce the most important techniques and provide examples that can be used across administration, teaching, and research.
Zero-shot prompting is the simplest form: you ask a question or give an instruction without providing the model with examples of the desired output. The model relies solely on its training.
- When it works: When the task is well known and unambiguous—for example translation, summarization, or simple questions.
- When it fails: When the output format is important, or when the task requires a specific style or tone.
Example
Simple prompt: What is the difference between GDPR and the Data Protection Act?
Why it works: The question is clear and well known. The AI can answer correctly without additional context.
Improved prompt: Explain the difference between GDPR and the Danish Data Protection Act. Focus on the practical differences that an employee at a Danish university needs to be aware of. Keep the explanation to a maximum of 200 words and avoid legal jargon.
Why it works: It adds target audience, format, length, and language style—still zero-shot, but much more precise.
With few-shot prompting, you provide the model with one or more examples of the desired input–output pattern before presenting your actual task. This is one of the most effective techniques for controlling format, tone, and quality.
Example
Classification prompt: Categorize the following student inquiries as either “IT support,” “Student administration,” or “Other.” Examples: “I can’t log in to DTU Learn” → IT support - “What is the deadline for registering for the exam?” → Student administration - “Where can I find the cafeteria?” → Other. Now categorize: “My printer doesn’t work on campus” - “Can I change my study programme after the first semester?” - “How do I get access to the VPN from home?”
Why it works: The three examples demonstrate the desired format. The model will respond consistently in the same style.
Writing style prompt: Write short news summaries for an internal newsletter in the following style: Input: “DTU has entered into a new collaboration with Novo Nordisk on research in sustainable production.” Output: “New collaboration: DTU and Novo Nordisk join forces on sustainable production research.” Input: “The IT department is upgrading all Wi‑Fi access points in buildings 303–305 in week 12.” Output: “Wi‑Fi upgrade: Buildings 303–305 get a new wireless network in week 12.” Now write a summary for: “The university has decided to introduce a new digital self-service solution for travel expense claims, replacing the current paper-based process from 1 April.”
Chain-of-Thought prompting asks the model to think step by step through a problem before providing its final answer. This technique significantly improves the quality of responses to complex problems, analyses, and decisions.
Example
Simple prompt: We are considering switching from local file servers to a cloud-based solution for document sharing in our department (approx. 80 employees). Think step by step through the advantages, disadvantages, and risks of this change. Consider finances, security, usability, and GDPR.
Why it works: “Think step by step” forces the model to systematically go through each perspective instead of giving a superficial answer.
Structured prompt: We need to organize a conference for 200 participants on campus in 3 months. Go through the planning step by step: 1. What needs to be clarified first? (venue, date, budget) 2. Which stakeholders need to be involved? 3. What are the typical pitfalls at this scale? 4. Propose a timeline with milestones 5. What is most often overlooked?
By assigning the AI a role, you define its expertise, tone, and perspective. This is particularly effective when you need responses from a specific professional viewpoint or in a specific communication style.
Example
Prompt, expert role: You are an experienced HR consultant specializing in onboarding at Danish universities. A department asks: “We are hiring three new employees at the same time next month. How do we ensure a good onboarding experience for all of them?” Respond as a practical advisor with concrete suggestions that can be implemented with limited resources.
Prompt, multi-role dialogue: Simulate a discussion between three perspectives on the question: “Should we introduce AI tools more broadly in our administrative processes?” 1. An IT security officer (focus on risks and data) 2. A department manager (focus on efficiency and time savings) 3. An employee (focus on job satisfaction and psychological safety) 4. Each person should contribute three statements. They should respond to one another.
This technique explicitly specifies the desired output format. It is particularly useful when the data will be reused in spreadsheets, presentations, emails, or other systems.
Example
Prompt, table format: Create a comparison table of the following video conferencing platforms: Zoom, Microsoft Teams, Google Meet, and Webex. Columns: Platform | Max participants | Recording | Screen sharing | Integrations | Price. Format it as a Markdown table. Include both the free version and paid licenses.
Prompt, structured email draft: Write an email to all employees in the department about planned downtime of the email system. Format: – Subject: [short and clear] – Introduction: [1 sentence about what is happening] – Timeframe: [when it will affect them] – What they need to do: [concrete actions] – Contact: [where to go in case of problems]. The tone should be professional but friendly. Keep the email under 100 words.