Claude Prompt Engineer — AI Landing Page System

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 🏢 WHO WE ARE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Digital marketing agency managing the website of an industrial manufacturer in Spain (energy sector, 80+ product landing pages). We use a Claude (Anthropic) project to generate landing pages from: ▸ An official HTML template ▸ A project system prompt ▸ Documented SEO rules ▸ Search Console data that sets priorities Pages are delivered as single self-contained HTML files pasted into WordPress (Custom HTML block, no JavaScript, Elementor full-width). Content is in Spanish — you don't need to speak it. Our team validates all Spanish output during iterations; you own the system. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 🚀 THIS IS A TRIAL FOR A LONG-TERM ROLE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ This project is a paid trial. We're an agency with multiple clients, and AI-powered systems like this one are core to how we work — landing pages, content pipelines, SEO automation, internal tools. If this project goes well, we want you as our go-to AI/prompt specialist on a recurring basis: new systems for new clients, improving existing ones, and advising the team. We're not looking for a one-project freelancer. We're looking for the right profile to grow with us. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ ⚠️ THE PROBLEM ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Output is inconsistent. In some runs the AI: ✗ Breaks our meta title/description rules, or overwrites metadata that must be preserved ✗ Drifts from the template architecture — section order, skipped blocks, altered styles ✗ Makes decisions it shouldn't — rewriting approved content, inventing technical specs ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 🎯 WHAT WE'RE HIRING ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ This is NOT a one-off audit. We want someone hands-on who: 1️⃣ REVIEWS our current setup (system prompt, template, SEO rules) and our WordPress→HTML workflow, and tells us what to fix 2️⃣ REBUILDS the project prompt — rule hierarchy, positive/negative examples, mandatory output format, self-validation checklist, "ask instead of deciding" protocol 3️⃣ WORKS WITH US ITERATIVELY — we generate real landings, we report what fails, you adjust, we test again… until landings come out correct WITHOUT manual fixes, every time 4️⃣ LOCKS THE ARCHITECTURE so we can scale — new landing types must follow the same structure with zero drift 5️⃣ MAKES IT OPERABLE BY ANYONE — documented so a non-technical junior can run it, replicate it for new landings, and maintain it without you ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ ✅ DEFINITION OF DONE — read carefully ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ The job is NOT complete when the prompt works for you. It is complete when: (a) 3 consecutive real landings generated by the system pass our full checklist (SEO metadata, template structure, styles, content rules) with ZERO manual corrections — AND — (b) a member of OUR team — junior, non-technical, without your help — follows your documentation, runs the system on a NEW landing, and gets a correct result on their own If it only works in your hands, it's not done. We're buying a replicable system, not a demo. Final milestone is only released when both conditions are met. This is also how we'll evaluate you for the long-term role: systems thinking, not heroics. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 🔧 MUST-HAVE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ ✔ Proven production experience with Claude/GPT system prompts (we'll ask for anonymized examples) ✔ Solid on-page SEO you can turn into machine-followable rules — metadata, headings, internal linking, preservation rules ✔ HTML/CSS to spot template deviations ✔ WordPress familiarity — Custom HTML, Elementor, caching ✔ Ability to document for non-technical operators ✔ Available for short iterative cycles over 2-4 weeks (async + quick reviews) — not a deliverable-and-disappear engagement ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ ⭐ NICE TO HAVE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ ▸ Spanish (content language — faster iterations, not required) ▸ Experience building QA checklists or eval suites for LLM outputs ▸ Search Console / Ahrefs familiarity ▸ Interest in a stable, recurring collaboration with an agency ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 🤝 HOW WE'LL WORK ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Fixed price · 3 milestones: ① Review + rebuilt prompt v1 ············ ~30% ② Iteration cycles ······················· ~40% ③ Replication test passed + docs ········· ~30% Access to all project docs and good/bad examples. Kickoff call 30-45 min, then async via Upwork. After a successful trial → recurring work on new AI systems for our other clients. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 📝 TO APPLY — answer these 3 questions (generic proposals will be discarded) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 1. A real case where a system prompt gave inconsistent outputs and how you fixed it — what changed, and how you measured the improvement. 2. Your specific technique to make an AI reliably PRESERVE an existing meta title instead of rewriting it. 3. What's wrong with this meta description for the keyword "silent generator", and your rewrite: "Quality electric generators at the best price. Come in and discover our wide range of products for all your needs." ⚡ Start your proposal with the word "TURBINE" so we know you read the full post.