Prompt Engineering Teardown: Three AI Prompts, One Clear Winner
Objective: Given three candidate AI writing prompts for the same marketing deliverable, evaluate each against the lesson's 4-part framework (Role, Task, Context, Constraints) to identify structural flaws, missing constraints, and hallucination risks before running them in production.
You're the content lead at Swiggy reviewing draft ChatGPT prompts created by junior marketers for Swiggy Instamart's monsoon flash-sale push notification and email campaign.
Analyze all three prompt specimens. For each prompt, identify missing framework components, vague instructions, or missing negative constraints (banned words) against the answer key.
Before you start
What you'll need
Free path (everything below is enough to finish)
Free tier is sufficient for testing prompt engineering variations
Free, structured rubric scoring
The process
Specimens to review
Evaluate this prompt against the 4-part structure (Role, Task, Context, Constraints). Name every defect you find.
Write a promotional email and 3 push notifications for Swiggy Instamart's monsoon campaign offering 20% off hot snacks and tea. Make it catchy, engaging, and exciting for foodies.
Specimen: synthetic, realistic
Evaluate this prompt against the 4-part structure. Name every defect you find, or state clearly that there are none.
You are a mobile direct-response copywriter for Swiggy Instamart in India. Task: Write 3 push notification variants (max 60 characters title, max 90 characters body) and one 80-word promotional email announcing 10-minute delivery of chai and pakoras during monsoon rains. Context: Target audience is urban working professionals ordering evening snacks between 4 PM and 7 PM. Tone: witty, relatable, conversational Indian-English. Past winning push example: 'Rain outside, steaming samosas at your door in 10 mins. 🌧️☕'. Constraints: Strictly observe character limits. Include exactly one clear CTA. Do NOT use: 'unlock', 'delve', 'seamless', 'game-changer', 'elevate your rainy day', or 'in today's fast-paced world'. Do not invent fake voucher codes.
Specimen: synthetic, realistic
Evaluate this prompt against the 4-part structure and production best practices. Name every defect you find.
You are an experienced copywriter for quick-commerce delivery apps. Write a 1,200-word comprehensive blog post about monsoon comfort food recipes and how Swiggy delivers ingredients in 10 minutes. Audience: home cooks and busy parents. Tone: warm and inspiring.
Specimen: synthetic, realistic
Final deliverable
A prompt scorecard evaluating all 3 specimens against the 4-part framework and a fully rewritten prompt for the flawed variants.
See a reference example
Prompt Scorecard, Freshworks Customer Support Campaign Specimen 1: 'Write a welcome email for Freshdesk users.' Role: No | Task: Vague | Context: No | Constraints: No Score: 0/4 pillars present — High risk of generic 5-paragraph AI fluff Specimen 2 (Clean Rewrite): Role: You are a senior lifecycle copywriter at Freshworks. Task: Write a 120-word welcome email for newly registered IT managers starting a 14-day Freshservice trial. Context: Audience values fast ticket resolution and SLA compliance. Match tone: direct, helpful, peer-to-peer. Constraints: Max 120 words. Include 1 primary CTA. Do not use: unlock, seamless, leverage, delve into, or game-changer. Score: 4/4 pillars present — Production ready
Success criteria
You're done when you can:
- Correctly identifies Specimen 2 as the clean 4-part prompt with zero defects
- Names specific missing prompt pillars (Role, Context examples, negative constraints) in Specimens 1 and 3
- Flags single-shot 1,200-word generation as an anti-pattern requiring the 3-pass method