Last month, I spent 3 hours trying to write a decent cold email template. Three. Whole. Hours. The AI kept spitting out generic garbage that sounded like every other “Hey [FIRST_NAME], hope this email finds you well”… Then I changed one thing in my prompt. One thing. Suddenly, the AI was writing emails that actually sounded human, referenced specific connection points, and had personality. My replies rates jumped tremendously! That moment? That’s when prompt engineering stopped feeling like a skill and started feeling like almost cheating. Wait, This Actually Works? Here’s the thing about prompt engineering that is obvious: it’s about getting really, really good at asking for exactly what you want. exactly Most of us suck at it. Because is not that easy. It clicked when I started to build this site using Cursor. Cursor My first attempts were disasters: My terrible prompt: My terrible prompt: "Create my homepage and style with stunning and aesthetic visuals" "Create my homepage and style with stunning and aesthetic visuals" The result: The result: Generic, ugly, messy code nobody would ever be able to customize. 🤮 My improved prompt: My improved prompt: "You're a senior web designer developer with deep knowledge in UI/UX. You are building my personal blog with me, a good fellow unfamiliar with our codebase (Astro Framework). Based in Astro conventions and best practices, create practical assets, components like UI and sections based in astro files. The final result should be a template that experienced developers could use and customize easily..." "You're a senior web designer developer with deep knowledge in UI/UX. You are building my personal blog with me, a good fellow unfamiliar with our codebase (Astro Framework). Based in Astro conventions and best practices, create practical assets, components like UI and sections based in astro files. The final result should be a template that experienced developers could use and customize easily..." The result: The result: Actually useful and clean astro files, at least better and more organized than before. (CSS files are still meeh though) 😅 The difference? I stopped asking the AI to write generic codes and started asking it to be an experienced developer helping a colleague building his humble project. The Five Rules That Changed Everything 1. Stop Being Polite to Robots I used to write prompts like I was asking a favor: “Could you please maybe help me write a blog post about SEO?”. Now I’m direct: “Write a 1,200-word blog post for marketing developers who want to understand technical SEO. Include code examples and explain why site speed actually matters for conversion rates, not just rankings.” The AI doesn’t have feelings. It has algorithms. Feed those algorithms exactly what they need. 2. Context is Everything Bad prompt: Bad prompt: "Write a LinkedIn post about growth marketing." "Write a LinkedIn post about growth marketing." Good prompt: Good prompt: "I'm a Marketing Engineer at a YC startup. Write a LinkedIn post sharing one specific growth hack I discovered while scaling our user base from 1K to 10K. Make it tactical, not theoretical. My audience is other growth marketers and technical founders." "I'm a Marketing Engineer at a YC startup. Write a LinkedIn post sharing one specific growth hack I discovered while scaling our user base from 1K to 10K. Make it tactical, not theoretical. My audience is other growth marketers and technical founders." The second prompt works because the AI knows: Who I am (Marketing Engineer, YC startup) What I want (specific growth hack) The context (1K to 10K scale) My audience (growth marketers, technical founders) The tone (tactical, not theoretical) Who I am (Marketing Engineer, YC startup) What I want (specific growth hack) The context (1K to 10K scale) My audience (growth marketers, technical founders) The tone (tactical, not theoretical) 3. Show, Don’t Just Tell Instead of saying “write in a conversational tone,” I give examples: "Write like this: Here's the thing nobody talks about with A/B testing: most marketers get so excited about statistical significance that they forget to check if the difference actually matters. I've seen teams celebrate a 2% lift on a metric that generates $50/month. Congrats, you just spent three weeks optimizing for an extra dollar a month" "Write like this: Here's the thing nobody talks about with A/B testing: most marketers get so excited about statistical significance that they forget to check if the difference actually matters. I've seen teams celebrate a 2% lift on a metric that generates $50/month. Congrats, you just spent three weeks optimizing for an extra dollar a month" The AI learns from the example and matches that specific style. 4. Constrain to Liberate Counterintuitive but true: the more constraints you give, the more creative the output. Instead of: Instead of: "Help me with marketing automation." "Help me with marketing automation." Try: Try: "I need a 7-email drip sequence for SaaS trial users who haven't logged in after day 3. Each email should be under 100 words, focus on one specific feature, include a clear and valuable CTA, sounding like it's coming from a helpful teammate, not a sales robot." "I need a 7-email drip sequence for SaaS trial users who haven't logged in after day 3. Each email should be under 100 words, focus on one specific feature, include a clear and valuable CTA, sounding like it's coming from a helpful teammate, not a sales robot." Constraints force creativity within boundaries. 5. Iterate Like Your Conversion Rate Depends on It My best prompts are never first drafts. I treat prompt engineering like optimizing ad copy (test, measure, refine, repeat). First attempt usually gets me 60% of what I want. Then I say: “Make it more specific” “Add a real example from the industry” “Remove the corporate buzzwords” “Include the technical details a developer would care about” “Make it more specific” “Add a real example from the industry” “Remove the corporate buzzwords” “Include the technical details a developer would care about” Each iteration gets closer to perfect. The “Cheating” Factor Here’s why prompt engineering feels like cheating: I’m getting expert-level outputs on topics I’m still learning about. expert-level outputs I needed to ship a free Astro template. Instead of spending hours reading documentation, I just: free Astro template Watched some videos to learn best conventions (playlist) Downloaded some important teachings transcripts Feeded AI with Astro Docs + useful tutorials Was really specific I wanted to make a clean code and replicable template Watched some videos to learn best conventions (playlist) playlist Downloaded some important teachings transcripts Feeded AI with Astro Docs + useful tutorials Was really specific I wanted to make a clean code and replicable template clean code replicable template The Marketing Engineer Advantage Here’s what I’ve learned being caught between marketing and engineering teams: both sides are already using AI, but they’re using it differently. Marketers use AI for content: social posts, email copy, blog outlines. Marketers Engineers use AI for code: debugging, documentation, optimization. Engineers As a Marketing Engineer, I am trying to use AI to translate between worlds: Marketing Engineer Converting technical features into benefit-driven copy Turning marketing requirements into technical specifications Building automation workflows to help both teams Converting technical features into benefit-driven copy Turning marketing requirements into technical specifications Building automation workflows to help both teams The prompt engineering skills transfer directly. Asking AI to debug a Python script or write an email sequence, it’s the same core skill: being incredibly specific about what I want. being incredibly specific about what I want The Real Superpower Prompt engineering isn’t actually about AI. It’s about getting incredibly good at articulating exactly what you want. That’s why I believe to be better we need to be learning, reading, and discovering something, always. And writing thoughts in somewhere. This is exactly how I built this blog, by applying prompt engineering to create content that ranks well and helps readers. And the specificity skill will transfer everywhere: Better briefs for your design team Clearer requirements for developers More effective communication with stakeholders Stronger positioning for your products Better briefs for your design team Clearer requirements for developers More effective communication with stakeholders Stronger positioning for your products So yes, prompting well feels like cheating. This is just the latest one. This is just the latest one. What’s your best prompting win? Want to see prompt engineering in action? Check out how I used these techniques to build this blog with perfect SEO scores and create content that ranks. Want to see prompt engineering in action? Check out how I used these techniques to build this blog with perfect SEO scores and create content that ranks. build this blog with perfect SEO scores
