The AI-Powered Academic: Prompting

Foundations: Mastering Prompt Engineering

Prompting is the practice of designing context, constraints and checks so that AI outputs are useful without being mistaken for evidence.

Download Prompt Templates

These templates are designed to be copied and pasted directly into a large language model. They provide a structure for common academic tasks.

The Structured Feedback template helps staff to transform unstructured notes into student-friendly, formative feedback.
The Socratic Tutor template provides a prompt for students, encouraging them to learn a topic through cooperative dialogue with the AI rather than focusing on pure output generation.
The Academic Reviewer template offers a detailed structure for generating a peer-review-style critique of a written document.

The 7-Part Prompt Framework

For research, teaching and feedback tasks, a good prompt should define the work and the verification route. This structure adapts the research guide’s framework for everyday academic use.

1. Role: define the expert lens or teaching stance.
2. Task: state the action: evaluate, draft, compare, extract or explain.
3. Context: give the audience, level, purpose and stage of work.
4. Evidence: provide the text, data or sources the model should use.
5. Constraints: forbid invented citations, unsupported claims and overclaiming.
6. Format: specify the output structure, length and headings.
7. Verification: ask the model to flag uncertainty, missing evidence and checks still needed.

The 10 Principles of Effective Prompting

Use these core techniques as practical starting points, then test outputs against the evidence and the task requirements.

1. Clarity & Specificity

Tell the LLM exactly what you want, leaving no room for interpretation. LLMs are extraordinarily creative, so vague prompts can lead to varied, inconsistent outputs.

Instead of: "Produce a report based on this data"
Use: "List our five most popular products and write a one-paragraph description of each"

2. Context Provision

Furnish all necessary background information. Context helps the LLM narrow its vast knowledge to your specific needs and avoid generic outputs.

Example: "I am a college senior with a 3.5 GPA and I need an essay outline on the French Revolution's impact"

3. Role/Persona Assignment

Assigning a specific role directly influences tone, style, and domain expertise. This makes responses more focused and professional.

Example: "You are a patent lawyer. Explain the legal process for patenting an invention in simple terms"

4. Output Format Definition

Clearly specify the desired structure for machine-readable outputs like JSON, XML, or human-readable formats like lists and tables.

Example: "Return results in JSON: {'key': 'value'}" or "Provide a concise summary in bulleted list format"

5. Examples (Few-Shot Prompting)

Including examples can lead to massive improvements in accuracy. Even a single example can significantly guide the model to desired output structure and style.

Example: "Here's an example of a one-paragraph description for another product..." then provide your request

6. Iterative Refinement

Prompt engineering is rarely "one-and-done." Continuously refining prompts based on LLM responses is essential for improving quality.

Example: Start broad, then refine: "Based on the outline provided, expand on the target audience section..."

7. Conciseness/Information Density

LLM performance can decrease with prompt length. Improve "information density" by shrinking information into fewer words.

Instead of: "The overarching aim of this exceptionally well-structured..."
Use: "Produce high quality, readable, clear content"

8. Chain of Thought (CoT) Prompting

For complex problems, encourage step-by-step reasoning. This enhances reasoning capabilities and provides transparency into the model's logic.

Example: "Let's think step-by-step" or "Explain each step" for mathematical problems

9. Instructions over Constraints

Instruct the model what to do (positive instructions) rather than what not to do (constraints). Use clear binary "hard on/off rules."

Instead of: "Do not list video game names"
Use: "Only discuss the console, company, year, and total sales"

10. Testing & Data-Driven Approach

Test prompts empirically using a "Monte Carlo approach"—generate multiple outputs and evaluate quality for statistical reliability.

Example: Use a spreadsheet to track "Prompt," "Output," and "Good Enough" ratings across 10-20 attempts

Related AI Literacy Sites

Move between the staff, research, student and beginner-facing guides. Product features, pricing, model access and privacy settings change frequently, so check current provider documentation before relying on a platform for teaching, research or sensitive work.