Prompt Engineering Examples — Real Before and After Prompts
The fastest way to learn prompt engineering is to see exactly what changes between a weak prompt and a strong one. Below are 6 real examples across different use cases — each with a clarity score, the specific improvements made, and the reasoning behind them.
Scores are generated by Deepclario's prompt analyzer — rated across goal clarity, context, format, constraints, and examples.
Weak prompt
“Write a product description”
Strong prompt
Write a product description for a $49/month project management tool aimed at freelance designers. The key benefit: it replaces 4 separate apps. Tone: confident and concise. Length: 3 sentences max. Avoid corporate jargon.
The original prompt has no audience, no product details, no tone guidance, and no length limit. Every one of those gaps gives the AI room to guess wrong.
Weak prompt
“Write a login function”
Strong prompt
Act as a senior Node.js developer. Write a login function using Express and bcrypt that: (1) accepts email and password, (2) checks against a PostgreSQL users table, (3) returns a signed JWT on success, (4) returns a 401 with a generic error message on failure to avoid user enumeration. Add inline comments explaining the security decisions.
Coding prompts fail when they lack the stack, the security requirements, and the output format. A junior developer's prompt and a senior developer's prompt look completely different.
Weak prompt
“Summarize this paper for me”
Strong prompt
Summarize this research paper in 3 clearly labeled sections: (1) Core hypothesis in one sentence, (2) Methodology in 2–3 sentences — what they measured and how, (3) Key finding and its practical implication for clinicians. Target reader: a hospital administrator with no research background.
A summary without a target audience is summaries for no one. Defining who will read it changes what gets included and what gets cut.
Weak prompt
“Write a cold email”
Strong prompt
Write a cold outreach email to a VP of Engineering at a 50–200 person SaaS company. Goal: get a 20-minute call to demo our code review tool. Their pain point: PRs are taking 3+ days to get reviewed. Keep the email under 100 words. No "I hope this finds you well." End with one specific question, not a generic CTA.
"Cold email" means nothing to an AI. Define the recipient, the goal, the pain point, the length limit, and what to avoid — and the output is actually usable.
Weak prompt
“Analyze my business idea”
Strong prompt
Act as a venture capitalist who has evaluated 500+ B2B SaaS companies. Analyze this business idea: [idea]. Structure your response as: (1) Market size estimate, (2) Top 3 risks, (3) Who the likely first 100 customers are, (4) One question you would ask before investing. Be direct and critical — don't soften feedback.
Open-ended analysis prompts produce open-ended answers. A structured output format with a critical role forces the model to take a position.
Weak prompt
“Explain quantum computing”
Strong prompt
Explain quantum computing to a 16-year-old who understands basic algebra but has never studied physics. Use one analogy involving everyday objects. Limit the explanation to 4 paragraphs. End with one question that would help them think deeper about the topic.
"Explain X" is one of the most common prompts and one of the least effective. It produces textbook-level output. Specifying the audience and asking for an analogy transforms the response.
Analyze your own prompts
Paste any prompt and see exactly which dimensions are weak — and get a rewritten version that scores higher. Free to try, no account needed.