Before/after prompt rewrites showing exactly what changes and why.
The fastest way to get better at prompting is to see concrete examples of what changes and why it works. Here are six transformations across common use cases.
Before: > "Write an email about the new pricing."
After: > "Write a professional email to our existing customers announcing a 15% price increase effective March 1st. Acknowledge that this is significant news. Briefly explain that the increase reflects rising infrastructure costs. Offer a 60-day lock-in at current pricing for customers who renew before March 1st. Keep it under 200 words. Tone: honest and direct, not corporate."
What changed: Added the audience, the specific details (15%, March 1st, 60-day offer), the reason for the change, length constraint, and tone direction. The first prompt produces a generic template. The second produces something close to sendable.
Before: > "Explain machine learning."
After: > "Explain machine learning to a non-technical marketing manager in under 200 words. Use one concrete analogy. Focus on what it actually does in practice, not how it works mathematically."
What changed: Added the specific audience (non-technical marketing manager), a length constraint, an instruction to use an analogy, and a directive about what to focus on vs. avoid. The before version produces a textbook overview. The after version produces something that person can actually use.
Before: > "Write a function to validate email addresses."
After: > "Write a Python function to validate email addresses. Requirements: (1) checks that the format is valid using regex, (2) returns True/False, (3) handles edge cases like missing @ or domain, (4) includes a docstring explaining parameters and return value. Include three test cases covering valid input, invalid format, and missing domain."
What changed: Added language, specific requirements, expected return format, edge case handling, documentation requirement, and test cases. The before version produces a basic regex check. The after version produces production-quality code with tests.
Before: > "Tell me about remote work."
After: > "I'm writing a 1,500-word article for managers at mid-size companies who are deciding whether to keep, reduce, or expand remote work policies. Give me: (1) the three strongest arguments for remote work backed by research findings, (2) the three most significant genuine downsides, (3) what the current evidence says about productivity impacts. Use balanced, evidence-based framing. Note where the evidence is contested."
What changed: Added the purpose, the specific audience, the exact structure wanted, a specific scope (productivity), and a framing directive (balanced, note where contested). The first prompt produces a surface-level overview of remote work. The second produces something structurally useful for writing the article.
Before: > "Make this better."
After: > "Edit the following paragraph for a technical blog aimed at senior software engineers. Goals: (1) cut it to under 100 words, (2) make the opening sentence stronger — it currently buries the key point, (3) remove any hedging language like 'might' or 'could potentially.' Do not change the technical content or examples.
[paste paragraph]"
What changed: Added audience, specific editing goals with clear priority, a word count target, a specific problem to fix (buried lede), and an explicit constraint (don't touch the technical content). "Make this better" leaves all judgment to the AI. The revised prompt is a specific editorial brief.
Before: > "Analyze our customer survey results."
After: > "Analyze the following customer survey results. Answer these specific questions: (1) What is the primary driver of dissatisfaction among users who rated us 3 or below? (2) Is there a pattern in the language used by promoters vs. detractors? (3) What one change do the results most strongly suggest we should prioritize?
Format your response as: Key Finding → Supporting Evidence → Recommended Action for each question.
[paste survey data]"
What changed: Replaced "analyze" (which is almost infinitely broad) with three specific questions to answer. Added a specific output format tied to each question. The before version produces a general summary. The after version produces an analysis structured around actual decisions.
Looking across all six examples, the improvements come from the same moves: - Adding audience and purpose so the AI knows who the output serves - Being specific about deliverables rather than naming a task category - Providing format structure so output can be used immediately - Adding constraints (length, what not to do, what not to change) - Breaking vague goals into concrete sub-questions
None of this requires expertise. It requires the habit of asking yourself: "What would I need to tell a capable person to get exactly what I want?" — then putting that in the prompt.
Have a follow-up question about this topic?
Ask AI