Writing structured prompts (templates)

Imagine pannunga — neenga oru builder kitta veedu kattanum nu solreenga.

Approach 1: "Oru nalla veedu kattunga" — Builder confused. Evlo rooms? Budget? Style? Yaar ku?

Approach 2: "3 BHK, modern style, ₹50L budget, 4-person family ku, east-facing, 1200 sq ft, modular kitchen venum" — Builder exactly enna venum nu theriyum!

AI prompting la um same dhaan. Structured prompts = detailed blueprint. Random prompts = vague instructions.

Previous article la Zero-shot, Few-shot, CoT learn panneenga. Now frameworks learn pannunga — these are SYSTEMS for writing consistently great prompts every single time.

Indha article la cover pannaporadhu:

• 🏗️ RICE Framework — 4-step prompt structure
• 🎨 CREATE Framework — 6-step advanced structure
• 🎭 Role-based prompting — AI ku roles assign pannunga
• 📋 5 ready-to-use templates — copy-paste and customize
• ⚡ Real before/after examples — difference feel pannunga

Indha article mudichaa, neenga "prompt templates" oru personal library maintain panna start pannuvenga. Let's build that library! 📚

Prompt framework = oru structured method to organize your instructions to AI. Random ah type pannradhu ku badhila, oru system follow pannreenga.

Why frameworks matter:

Without framework:

*"Write me a blog post about AI"*

→ Generic, unfocused, probably 500 words of fluff

With framework:

*"Role: Tech blogger for Tamil audience. Task: Write a 300-word blog post about how AI is changing Indian IT jobs. Tone: Conversational, optimistic. Format: 3 sections with headers. Include: 2 statistics, 1 real company example."*

→ Focused, specific, exactly what you want!

The 2 main frameworks:

RICE Framework (Simple, 4 components):

• Role — Who should the AI be?
• Instruction — What should it do?
• Context — Background information
• Expectation — What output format/quality you expect

CREATE Framework (Advanced, 6 components):

• Character — AI's persona/expertise
• Request — Specific task
• Examples — Few-shot examples (optional)
• Adjustments — Constraints, tone, length
• Type — Output format (list, essay, code, table)
• Extras — Additional requirements

Which one to use?

RICE simple but powerful. Daily tasks la idhu podhum.

R — Role (யார் மாதிரி?):

AI ku oru persona assign pannunga. Idhu output quality dramatically change pannudhu.

Examples:

• *"You are a senior marketing manager with 10 years experience"*
• *"You are a Tamil teacher explaining to 10-year-old students"*
• *"You are a startup CTO evaluating technology choices"*

I — Instruction (என்ன செய்யணும்?):

Clear, specific task. Vague instruction = vague output.

❌ "Write about marketing"

✅ "Write 5 Instagram caption ideas for a new coffee shop launch in Chennai"

C — Context (Background என்ன?):

Relevant information AI ku theriyaadha or remember pannanum.

• Target audience details
• Industry/domain specifics
• Previous related work
• Constraints (budget, timeline, etc.)

E — Expectation (Output எப்படி வேணும்?):

Exact format, length, tone, style specify pannunga.

• *"Output as a markdown table with 3 columns"*
• *"Keep it under 200 words, professional tone"*
• *"Include code examples in Python 3.11"*

Full RICE Example:

*"Role: You are a senior HR manager at a mid-size IT company in Chennai.*

*Instruction: Draft a rejection email to a candidate who reached the final round.*

*Context: The candidate, Priya, was strong technically but lacked leadership experience. We want her to reapply in 6 months.*

*Expectation: Professional, empathetic tone. Under 150 words. Include specific feedback and encouragement to reapply."*

Idhu random ah "write a rejection email" nu type pannaa varadhu la — 10x better output varum! 🎯

CREATE RICE oda big brother maari. Complex tasks ku more control tharum.

C — Character (Persona + Expertise):

RICE oda Role maari, but more detailed. Personality, expertise level, communication style ellam specify pannalaam.

*"You are Dr. Ramesh, a data science professor at IIT Madras with 15 years experience. You explain complex concepts using Tamil analogies and real Indian examples."*

R — Request (Specific Task):

What exactly you want. Multiple sub-tasks irundaa, numbered list la kudungo.

*"1. Explain what Random Forest is. 2. Compare with Decision Tree. 3. Give a real-world use case from Indian banking."*

E — Examples (Optional Few-shot):

Pattern establish pannanum na, examples add pannunga.

*"Like this style: 'Random Forest enna na, imagine pannunga 100 doctors oru patient ah examine pannuraanga...'"*

A — Adjustments (Constraints & Tone):

Length, tone, what to include/exclude, language mix.

*"Tone: Casual, conversational Tanglish. Length: 300-400 words. Avoid: Complex math formulas. Include: At least 2 analogies."*

T — Type (Output Format):

Exact format specify pannunga.

*"Format: Blog post with H2 headers, bullet points for key concepts, a comparison table, and a summary box at the end."*

E — Extras (Additional Requirements):

SEO keywords, call-to-action, references, anything else.

*"Include SEO keywords: 'machine learning Tamil', 'random forest explained'. End with 3 discussion questions."*

When CREATE > RICE:

Role-based prompting = AI ku oru specific identity assign pannunga. Idhu surprisingly powerful technique.

Why roles work:

LLM training data la experts oda content irukku — doctors, lawyers, teachers, developers. Role assign pannaa, AI adha category oda knowledge and communication style activate pannudhu.

Role writing formula:

*"You are a [expertise level] [profession] with [X years] experience in [specific domain]. You [key characteristic]. Your audience is [target]."*

10 Powerful Roles:

Role stacking — Multiple roles combine pannalaam:

*"You are a senior developer WHO ALSO has experience as a technical writer. Your code explanations are clear enough for junior developers to understand."*

Common mistake: Over-specific roles that don't exist in training data.

❌ *"You are a blockchain-powered quantum-AI specialist for Tamil Nadu agriculture"*

✅ *"You are an agricultural technology consultant familiar with Indian farming practices"*

Pro tip: Role description la negative instructions um add pannunga:

*"You are a financial advisor. You do NOT give specific stock recommendations. You focus on general financial literacy and principles."*

🎬 Movie director analogy:

Unstructured prompt = Director solluvaaru: "Act pannu." Actor confused — comedy aa? tragedy aa? love scene aa?

RICE prompt = Director solluvaaru: "Nee oru father (Role). Un daughter marriage fix aagirukku, un emotions show pannu (Instruction). She's marrying someone from another city, you're happy but sad she's leaving (Context). Subtle ah, overacting vendaam, tears varaama control pannu (Expectation)."

CREATE prompt = Director solluvaaru everything above + "Previous scene la nee laughing ah iruntha (Example). Camera close-up la irukku so micro-expressions matter (Adjustment). Dialogue illa, expression only (Type). Background music slow violin irukku, pace match pannu (Extras)."

Same actor, same scene — but structured direction gives Oscar-worthy performance vs random acting!

Unnoda AI um ipdi dhaan — better instructions = better performance. Framework use pannunga, results feel pannunga! 🏆

Under the hood, structured prompts epdi help pannudhu nu paapom.

1. Attention Mechanism Optimization:

LLM attention mechanism specific tokens la focus pannudhu. Structured prompt la clear sections irukku — Role, Task, Format — so attention correctly distributed aagum. Vague prompt la attention scattered, output unfocused.

2. Context Window Efficiency:

Every token counts. Structured prompts waste tokens avoid pannudhu — no repeated information, no ambiguity that needs clarification. Same context window la more useful information fit aagum.

3. Output Anchoring:

Format specification ("as a table", "in 3 bullet points") AI oda output generation ku strong anchor kudukudhu. Without it, AI default format use pannudhu — which may not be what you want.

4. Role Priming:

Role assignment first tokens la varuvadhaal, subsequent generation ellam adha role oda "lens" through filter aagum. "You are a doctor" sonna, medical terminology, cautious language, evidence-based reasoning automatically increase aagum.

Before vs After comparison:

Key insight: Structured prompts la neenga 5 min extra spend pannreenga prompt writing la. But 3-4 iterations save aagum. Net time savings: 60-70%!

Oru professional always tools use pannuvaaru — carpenter ruler use pannuvaaru, doctor stethoscope use pannuvaaru. Unnoda tool = prompt framework. Use it! 🛠️

Different roles ku different template usage:

👨‍💻 Software Developer:

📊 Marketing Professional:

🎓 Student:

👔 HR Professional:

Pro tip: Unnoda role ku relevant templates oru document la save pannunga. Reuse, refine, repeat! 📁

⚠️ Frameworks use pannumbodhu avoid pannunga:

Over-engineering:

- ❌ Simple question ku CREATE full framework: "What's 2+2?" ku 6-part prompt — overkill!

- ✅ Simple tasks ku simple prompts. Framework complex tasks ku dhaan.

Conflicting instructions:

- ❌ "Be concise" + "Include detailed examples with explanations" — AI confused aagum

- ✅ Pick one priority. Concise OR detailed, not both.

Role hallucination:

- ❌ "You are a doctor" → AI gives medical advice it shouldn't

- ✅ Always add: "This is for educational purposes. Recommend consulting a real professional."

Template rigidity:

- ❌ Same template for every task without modification

- ✅ Templates are starting points — customize for each specific use case

Missing context:

- ❌ Great structure but no relevant context: RICE without the C

- ✅ Context is often the most important part — don't skip it!

Format overload:

- ❌ "Output as a table with headers, then a bullet list, then a paragraph summary, then a flowchart, then..."

- ✅ One primary format. Maybe one secondary. Keep it focused.

Language mismatch:

- ❌ English framework for Tamil output without specifying language

- ✅ Explicitly state: "Respond in Tanglish (Tamil + English mix)"

Remember: Frameworks are tools, not rules. Break them when it makes sense! 🔧

Structured prompting oru skill alla — oru competitive advantage.

The productivity multiplier:

Average knowledge worker AI use pannumbodhu 3-4 iterations venum to get good output. Structured prompt use pannaa? First try la 80%+ quality. Over a week, indha time savings massive.

Team impact:

• Prompt templates share pannalaam across team
• New team members faster onboard aagum
• Output quality standardized aagum
• Best practices documented and improved over time

Career advantages:

1. Efficiency — Same work, less time, better quality
2. Leadership — You can teach others, become the AI expert
3. Innovation — Better prompts = better AI outputs = better ideas
4. Documentation — Your prompt library is intellectual property

Industry demand:

Companies now hiring "Prompt Engineers" — but really, EVERY role needs structured prompting skills. Marketing, engineering, HR, sales — ellam AI-assisted workflows adopt pannudhu.

Your action item: Start building a personal Prompt Library today. Notion, Google Doc, or simple text file — doesn't matter. Save your best prompts. Categorize by task type. Refine over time. In 6 months, you'll have a powerful toolkit that makes you 3x more productive! 🚀

📌 5 Things to Remember:

1. RICE = Simple tasks. Role + Instruction + Context + Expectation. Daily emails, quick content, code tasks — RICE podhum. 30 seconds la framework setup pannalaam.

1. CREATE = Complex tasks. Character + Request + Examples + Adjustments + Type + Extras. Reports, campaigns, documentation — CREATE use pannunga. Extra 2 min investment = dramatically better output.

1. Role-based prompting is powerful. "You are a senior [role]" add pannaa, output quality noticeably improve aagum. Roles AI oda relevant knowledge activate pannudhu.

1. Templates save time. 5 core templates maintain pannunga. Reuse and refine. First-try success rate 30% → 80% pogum.

1. Don't over-engineer. Simple task ku simple prompt. Framework complex tasks ku dhaan. Right tool for right job — previous article la learn pannadha remember pannunga!

Quick decision guide:

🎯 Challenge: Create your personal prompt template library!

Step 1 (5 min): Pick your top 3 most common AI tasks. Examples:

• Writing emails
• Summarizing documents
• Generating social media posts
• Debugging code
• Explaining concepts

Step 2 (10 min): For each task, write a RICE template:

• Fill in Role, Instruction, Context, Expectation
• Leave placeholders for variable parts: [topic], [audience], [length]

Step 3 (5 min): Test one template on ChatGPT/Gemini:

• Fill in the placeholders with real values
• Compare output with your usual unstructured prompt
• Note the quality difference!

Step 4 (Ongoing): Save your templates somewhere accessible:

• Notion page
• Google Doc
• Notes app
• Even a text file on desktop!

Bonus Challenge: 🔥

Take the CREATE framework and write a template for the MOST COMPLEX task you do at work. Share it with a colleague. Get their feedback. Refine it.

Expected outcome: After this challenge, you'll have 3 reusable templates that save you 15-20 minutes every day. Over a month, that's 8+ hours saved! Worth the 20 min investment, right? 💡

🎤 Prompt engineering interview preparation:

Q1: "What is a prompt framework and why is it useful?"

A: A prompt framework is a structured method for organizing AI instructions — like RICE (Role, Instruction, Context, Expectation) or CREATE (Character, Request, Examples, Adjustments, Type, Extras). It improves output quality, ensures consistency, enables reusability, and reduces iteration cycles from 4-5 to 1-2.

Q2: "How would you design a prompt template for a team to use?"

A: Start with the most common team tasks. Use CREATE framework for complex tasks. Include clear placeholders with instructions (e.g., [INSERT TARGET AUDIENCE — age, location, interests]). Add example filled templates. Version control the templates. Gather feedback and iterate monthly.

Q3: "What's the difference between role prompting and system prompting?"

A: Role prompting is done in the user message ("You are a senior developer..."). System prompting uses the system message field (available in API). System prompts are stronger — they persist across the conversation. In practice, for single-turn tasks, role prompting in user message works well. For multi-turn applications, system prompts are preferred.

Q4: "How do you handle conflicting requirements in a prompt?"

A: Prioritize requirements explicitly. Use numbered priority: "Priority 1: Accuracy. Priority 2: Brevity. If conflict, accuracy wins." Alternatively, split into multiple prompts — one for each requirement — then combine results.

Q5: "Can you give an example of a bad structured prompt and how to fix it?"

A: Bad: "You are an expert. Write something good about AI. Make it professional but casual. Long but concise." — contradictions everywhere! Fix: "You are a tech journalist. Write a 300-word LinkedIn post about AI in healthcare. Professional tone with one personal anecdote. Include 2 statistics." — clear, non-contradictory, specific.

🌟 The real secret of prompt engineering:

Idhu about AI illa — idhu about clear thinking. Structured prompt ezhudha therinja, neenga actually clear ah think panna therinjirukka nu artham.

RICE or CREATE framework fill pannumbodhu, neenga force pannreenga yourself to answer: "Who am I talking to? What exactly do I want? What context matters? What does good output look like?"

Indha questions answer panna therinja, AI mattum illa — email writing, presentation, meeting communication — ellam improve aagum.

Start today: Pick ONE template from this article. Use it for your next AI task. Feel the difference. Then build your library, one template at a time.

Frameworks are not constraints — they are launchpads. 🚀

🗺️ Your journey continues:

✅ Completed: Prompt types (Zero-shot, Few-shot, CoT)

✅ Completed: Structured prompts (RICE, CREATE, templates)

📍 Next: AI Tools Ecosystem — text, image, video tools comparison

🔮 Coming up: AI Hallucination, Using AI for Daily Work

Practice plan:

1. Today — Create 3 RICE templates for your common tasks
2. This week — Try CREATE for one complex task
3. This week — Build a "Prompt Library" document
4. Next week — Share templates with a colleague, get feedback

Your prompt library is your superpower. Every template you create makes you faster, better, more consistent. Invest time now, save hours later.

Keep building, keep refining! 📚