Back to Templates

Generate and Auto-Evaluate Facebook Ad Headlines using GPT-4o-mini

Created by

Created by: Yaron Been || yaron-nofluff

Yaron Been

Last update

Last update 5 hours ago

Share


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.