In 20 years of AI research and teaching, many technologies have promised to revolutionise higher education teaching and learning, but few have been adopted as rapidly as generative artificial intelligence (GenAI). While usage is widespread at my university, literacy levels remain inconsistent.
The primary issue is “cognitive noise”. Students write poorly constructed, vague prompts that result in generic or shallow outputs. To bridge this gap, we must prevent students from treating prompting as a shortcut and start teaching the structured design of GenAI interaction.
A model for GenAI prompt design
Most students treat AI like a search engine, asking simple, one-line questions. To deepen engagement, we teach the intent-context-constraint (ICC) model. Instead of a weak prompt like “Explain data mining”, students learn to build a structure.
- Intent: the specific learning goal (for example, “Explain data mining to a second-year student”)
- Context: necessary background (for example, “Include a real-world example”)
- Constraint: the required form (for example, three short paragraphs).
In our workshops, students using this structured approach produce significantly more coherent and discipline-aligned outputs compared with those using open-ended queries.
Build criticality through continued interaction
A major barrier to learning is the habit of asking a GenAI tool a question and accepting the first answer uncritically. We teach students that AI interaction should be a cycle, using the following framework: generate – refine – challenge – extend.
Encourage students to follow a generated result with: “Now simplify this for a non-technical audience” or “Identify the limitations of this argument”. This iterative process mirrors how expert thinking develops: progressively and critically, rather than instantly.
- Spotlight guide: Critical thinking in teaching and research
- Spotlight guide: Bringing GenAI into the university classroom
- AI as tutor and critic - using tech to personalise education
Reduce ambiguity with the 2+1 constraint
The 2+1 constraint is a precision tool designed to eliminate “cognitive noise” and generic outputs. While ICC identifies what to include, the 2+1 rule dictates how much detail is required by enforcing a formula: two specifications (such as a target audience and an example) plus one structural limit (such as a word count). For instance, “Explain blockchain to a beginner, include one real-world example, and limit the answer to 120 words”.
Use role-based prompting to simulate expertise
AI outputs improve dramatically when the tool is assigned a specific persona, a method known as “role framing”. Asking the AI to act as a critical reviewer, a university lecturer helps students generate explanations that meet higher academic standards. We found this significantly improved the analytical depth of student-generated materials.
Shift from answer-seeking to thinking support
The greatest risk of AI is that it becomes a tool to bypass the struggle of thinking. To counter this, we apply the “cognitive shift rule”: prompts should guide the process, not just provide the result.
A powerful guided prompt looks like this: “Guide me through solving this engineering problem step-by-step, but pause before giving the final answer to let me try it myself.” When our students used such prompts they were better equipped to solve similar problems independently.
Make the process visible in assessment
If we do not assess the prompting process, students may not take the skill seriously. I recommend the “process transparency model” for coursework:
- Require students to submit their “prompt logs” alongside final assignments
- Ask them to reflect on how their prompts evolved based on the AI’s responses
- Evaluate both the final output and the critical thinking shown in the prompt design.
Embed verification as a mandatory step
Because AI is designed to be fluent, it often sounds confident even when its outputs are inaccurate. This means every prompt must include a verification step. For example, “Provide supporting sources for this information and identify any flaws in your explanation.” This protects academic integrity and builds fact-checking skills.
To foster deep learning, students need to move beyond asking AI for answers to designing prompts that improve critical thinking.
When we teach students to design their interactions, AI stops being a shortcut and becomes a catalyst for genuine intellectual growth.
AI Disclosure: This article was developed by the author with AI assistance for structural editing and alignment with Campus guidelines. All technical insights, metrics and strategies reflect the author’s professional expertise and peer-reviewed research.
Mogeeb A. A. Mosleh is a professor of artificial intelligence at the University of Science and Technology, Yemen.
If you would like advice and insight from academics and university staff delivered direct to your inbox each week, sign up for the Campus newsletter.
comment