Prompt Patterns — Role, Context, Constraints, Examples

In the previous lesson, you learned the basics of writing clear prompts. Now we are going to level up. This lesson covers structured prompt patterns — specific techniques that give you fine-grained control over AI output.

Think of these patterns as tools in a toolkit. You do not need to use all of them in every prompt. But knowing what is available means you can reach for the right tool when you need it.

---

Pattern 1: Role Prompting (System/Role Instructions)

One of the most powerful ways to shape AI output is to assign the model a role. When you tell the model who it "is," you activate a cluster of patterns in its training data associated with that role's expertise, vocabulary, and communication style.

How It Works

"You are an experienced secondary school physics teacher. Explain Newton's three laws of motion to a Year 9 class. Use everyday examples and avoid mathematical notation."

By assigning the role of "secondary school physics teacher," you are telling the model to:

• Use age-appropriate language
• Prioritise clarity over technical precision
• Include relatable examples
• Adopt an encouraging, educational tone

More Examples

Role	Effect on Output
"You are a professional copywriter"	Punchier, more persuasive language
"You are a patient maths tutor"	Step-by-step explanations, encouraging tone
"You are a senior software engineer"	Technical depth, best practices, production-quality code
"You are a sympathetic career advisor"	Warm, supportive, practical guidance
"You are a strict academic peer reviewer"	Critical, detailed, evidence-focused feedback

When to Use It

Role prompting is especially useful when:

• You need a specific tone or communication style
• You want expertise pitched at a particular level
• You want the model to adopt a particular perspective for analysis
• You are creating content that should sound like it comes from a specific type of professional

A Subtle But Important Point

The model is not actually becoming a physics teacher or a lawyer. It is activating patterns associated with how those professionals communicate in its training data. This means role prompting works best for roles that are well-represented in the training data (teachers, writers, developers) and less reliably for highly niche roles.

---

Pattern 2: Context Loading

Context is the background information the model needs to give you a relevant, useful response. The more relevant context you provide, the better the output.

Types of Context

Task context — What are you trying to accomplish?

"I am writing a business plan for a small bakery and need help with the financial projections section."

Audience context — Who will read or use the output?

"The audience is the school's board of governors, who are non-technical but senior decision makers."

Background context — What has already happened or been decided?

"We have already chosen React for the frontend and PostgreSQL for the database. I need help designing the API layer."

Constraint context — What limitations exist?

"The budget is limited to £5,000 and we need to launch within 3 months."

Pasting Source Material

One of the most powerful uses of context is pasting the actual text you want the model to work with. Instead of describing what you need, give the model the raw material:

"Here is a research paper abstract. Summarise the key findings in three bullet points suitable for a non-specialist audience:

[paste the abstract here]"

This works brilliantly for:

• Summarising articles or papers
• Editing or proofreading text you have written
• Extracting specific information from long documents
• Translating text
• Comparing multiple sources

---

Pattern 3: Constraints

Constraints are explicit boundaries you set on the output. They prevent the model from being too verbose, going off-topic, or using an unwanted style.

Common Constraint Types

Length constraints:

• "Keep your response under 200 words"
• "Write exactly 3 paragraphs"
• "Give me a one-sentence answer"

Format constraints:

• "Respond in bullet points only"
• "Format as a markdown table"
• "Use numbered steps"
• "Output valid JSON"

Content constraints:

• "Only discuss the economic impacts, not the social ones"
• "Focus on the period between 1900 and 1950"
• "Use only information from the text I provided — do not add outside knowledge"

Style constraints:

• "Write in British English"
• "Use a formal academic tone"
• "Avoid idioms and colloquialisms — this will be translated into other languages"

Audience constraints:

• "Suitable for a 10-year-old"
• "Written for an expert audience familiar with the terminology"

Why Constraints Matter