Generate and auto-evaluate Facebook ad headlines using GPT-4o-mini

Generate and Auto-Evaluate Facebook Ad Headlines using GPT-4o-mini Built with n8n + OpenAI This workflow captures a product description, generates ad headlines, evaluates them with custom criteria, decides whether another draft is needed, and finally sends the best version via Gmail. --- ⚡ Section 1: Capture the Brief & Build the Prompt 📝 FormTrigger\_CopywritingBrief → A simple form asks: “What is your product about?”* ⚙️ Set\_PromptForHeadline → Prepares the input by appending the instruction: “Write a Facebook ad headline for this product:” Benefit: Ensures consistent, structured prompts so the AI receives clear context every time. --- ✍️ Section 2: Draft the Headline 💬 LLM\_HeadlineWriterModel → GPT-4o-mini model provides the intelligence. ✍️ Agent\_HeadlineWriter → Generates a first-pass Facebook ad headline. Benefit: Produces creative copy instantly without waiting on a human writer. --- 📋 Section 3: Define Scoring Criteria 💬 LLM\_EvalCriteriaModel → Calls GPT-4o-mini again. 📑 Agent\_EvalCriteriaBuilder → Suggests 5 scoring parameters (scale 1-10). Example: Clarity, Relevance, Hook Strength, Brand Voice, Scroll-Stoppage. Benefit: Builds an objective, repeatable evaluation rubric automatically. --- 🔍 Section 4: Evaluate the Headline 💬 LLM\_HeadlineEvaluatorModel → Supplies reasoning power. 🔍 Agent\_HeadlineEvaluator → Applies the 5 criteria to the generated headline and outputs: JSON with scores per parameter An average score A plain-language bottom-line Benefit: Turns subjective copy quality into measurable numbers. --- 🔄 Section 5: Decide & Iterate (if needed) 💬 LLM\_BottomLineModel → Interprets the evaluation results. 🤔 Agent\_IterationDecision → Decides: Return NO → headline is acceptable. Return YES + feedback → headline should be rewritten. 🔀 If\_NeedMoreIterations → Branches: If NO → continue workflow. If YES → (loop wiring possible) headline can be regenerated with feedback. Benefit: Keeps iterating until the AI headline meets your standards. --- 📩 Section 6: Deliver the Result 📧 Send a message (Gmail node) → Sends the accepted headline via email. Benefit: Automates delivery of the polished, AI-approved headline to your inbox or team. --- 📊 Workflow Overview | Section | Purpose | Key Nodes | Benefit | | -------------------- | ---------------------------------- | ----------------------------------------------------- | ------------------------------ | | ⚡ Capture Brief | Collect product info & prep prompt | FormTrigger, Set | Structured AI input | | ✍️ Draft Headline | Generate first headline | LLM\HeadlineWriterModel, Agent\HeadlineWriter | Instant creative draft | | 📋 Define Criteria | Build scoring rubric | LLM\EvalCriteriaModel, Agent\EvalCriteriaBuilder | Objective evaluation | | 🔍 Evaluate Headline | Score headline & summarize | LLM\HeadlineEvaluatorModel, Agent\HeadlineEvaluator | Transparent quality check | | 🔄 Decide & Iterate | Accept or refine headline | LLM\BottomLineModel, Agent\IterationDecision, If | Only good results move forward | | 📩 Deliver Result | Share the final copy | Gmail | Automates delivery | --- ✅ Final Benefits 🚀 One-click workflow: from product description to tested headline. 📊 Automatic rubric: objective scoring each time. 🔄 Self-improving: poor headlines can auto-iterate with feedback. 📧 Direct integration: approved headlines land in Gmail instantly. 🧩 Fully modular: easy to extend with Google Sheets, Slack, or CRM nodes. ---

By Yaron Been

471