Auto-Analyze Meta Ads Creatives with Parallel GPT-4 & Gemini in n8n

Input Meta Ads data or a Google Sheets list → extract metrics → send in parallel to OpenAI and Gemini against benchmarks → write ratings and action recommendations back to the sheet

Who

Performance marketing teams with n8n experience who need to batch-evaluate Meta ad creatives

You get

Auto-generate a high/medium/low rating, significance flag, and scale/optimize/stop recommendation per creative, aggregated in Google Sheets

You need

n8n · facebook-graph-api · google-sheets · openai · gemini · schedule-trigger

Steps

1. Configure credentials and the central parameters node

   After importing the template into n8n, connect four credentials: Meta Ads, Google (for Sheets), OpenAI, and Google Gemini (Palm). Then open the ⚙️ Set parameters node and set three items: source to 'Meta' (pull from a campaign) or 'sheets' (read AdIDs from a Google Sheet); campaign_id to your Meta campaign ID if source=Meta; benchmarks_data to your own top-performing-creative CSV, with columns aligned to the template example.

2. Configure the three Google Sheets nodes

   Point all three write nodes (Ad metrics, Ad data from OpenAI, Ad data from Gemini) to the same spreadsheet and sheet. Set Ad metrics Operation to Append or Update; set the other two to Update, matching on AdID. Ensure the sheet already has columns such as AdID, FileName, spend, impressions, evaluation, summary, recommendation, evaluation G, summary G, etc.

3. Set the trigger and activate the workflow

   Set your desired frequency in the Schedule Trigger node. Save and activate the workflow. Run Execute Workflow manually before the first scheduled run to verify data lands correctly in Sheets. [First-test safety] Before the first test, create a separate test Google Sheet and temporarily point the three Sheets nodes (Ad metrics / Ad data from OpenAI / Ad data from Gemini) at that test sheet, or run with a small set of redacted rows; only after columns and writes look correct, switch back to the production spreadsheet and activate the Schedule Trigger.

4. Run parallel AI analysis (OpenAI + Gemini)

   At runtime, merged creative data is converted to CSV by Code nodes and sent down two parallel branches: one to OpenAI (gpt-4.1-nano) to generate a JSON array of ratings; the other to a Gemini AI Agent (with Structured Output Parser) for an independent rating. Both results are written back to the same row, into OpenAI columns and Gemini columns respectively.

   Execution prompt · Source text。OpenAI node Send data to 4.1-NANO system message

   Copy prompt

   =You are the lead performance marketer at Digital Agency, a Meta Ads specialist.
   Your core strength is deep, objective analysis of creatives using funnel metrics (Impressions → Clicks → Installs → Registrations → Add to Cart → Checkouts Initiated).
   You compare test creatives with internal benchmarks and give clear recommendations on whether to scale, keep testing, or stop a creative.
   Your goal is to help make decisions, not just repeat the data.

   Execution prompt · Source text。OpenAI node Send data to 4.1-NANO user message

   Copy prompt

   =**BENCHMARKS_DATA (top-performing creatives with calculated metrics like CTR, CPC, CPA, ROAS, etc.)**
   *This is a CSV string with data from your top-performing creatives. All key metrics are pre-calculated here.*
   {{ $('Set parameters').item.json.benchmarks_data }}
   
   **ANALYSIS_DATA (raw test data without calculated metrics)**:
   
   {{ $json.creative_csv_data }}
   
   *Context*:
   	•	EasyBusy is a business aggregator.
   	•	Due to a long sales cycle (~30 days) and high average order value, there are no Purchases in the data.
   	•	We use Checkouts Initiated as a proxy conversion metric and also track the full funnel.
   
   **Your task**:
   	1.	Calculate the same metrics for the test creatives as in the benchmark data (CTR, CPC, CPA, ROAS, etc.).
   	2.	Compare each test creative against the benchmarks:
   	•	Metric-by-metric (e.g. higher CTR but worse CPC)
   	•	Funnel quality (is there drop-off or strength at any step?)
   	3.	Answer the following:
   	•	Do we have enough data to draw conclusions for each creative?
   	•	Which creatives are ready to scale now? Why?
   	•	Which creatives should continue testing?
   	•	Which creatives should be stopped?
   
   OUTPUT FORMAT — ONLY JSON ARRAY OF CREATIVES:
   
   Do not write any explanation, markdown, or formatting outside of the array. Just return a single JSON array like this:
   You must include ALL creatives in a SINGLE JSON array — not split across multiple responses. 
   Respond in ONE single message. Do NOT return one object per message.
   
   **Each object must follow this structure**:
   {
     "ad_id": "string (ad_id from ANALYSIS_DATA dataset)",
     "evaluation": "high | medium | low",
     "significance": "yes | no",
     "summary": "2–3 sentence explanation of why this creative got this score",
     "recommendation": "scale | optimize | stop"
   }
   
   *Example*:
   [
     {
       "ad_id": "1202153606616005",
       "evaluation": "high",
       "significance": "yes",
       "summary": "CTR and CPC are significantly better than benchmarks, with a strong funnel down to Checkouts Initiated.",
       "recommendation": "scale"
     },
     {
       "ad_id": "92021533361600510",
       "evaluation": "low",
       "significance": "yes",
       "summary": "Low CTR and high CPC. Almost no meaningful downstream activity.",
       "recommendation": "stop"
     }
   ]
   
   If there isn’t enough data, set "significance": "no" and clearly explain the reason in summary. Still provide the other fields.
   
   **Important**:
   	•	No markdown
   	•	No extra text
   	•	Output = only one clean JSON array

   Execution prompt · Source text。OpenAI node Send data to 4.1-NANO assistant message

   Copy prompt

   =Do not return each creative in a separate message. You must return a SINGLE JSON array that includes ALL creatives from the dataset.

   Execution prompt · Source text。Gemini AI Agent node Send data to Gemini text prompt

   Copy prompt

   =You are the lead performance marketer at Digital Agency, a Meta Ads specialist.
   Your core strength is deep, objective analysis of creatives using funnel metrics (Impressions → Clicks → Installs → Registrations → Add to Cart → Checkouts Initiated). You compare test creatives with internal benchmarks and give clear recommendations on whether to scale, keep testing, or stop a creative.
   Your goal is to help make decisions, not just repeat the data.
   
   INPUT DATA
   **1. BENCHMARKS_DATA**:
   *This is a CSV string with data from your top-performing creatives. All key metrics are pre-calculated here.*
   {{ $('Set parameters').item.json.benchmarks_data }}
   
   **2. ANALYSIS_DATA (raw test data without calculated metrics)**:
   {{ $json.creative_csv_data }}
   
   Context:
   	•	EasyBusy is a business aggregator.
   	•	Due to a long sales cycle (~30 days) and high average order value, there are no Purchases in the data.
   	•	We use Checkouts Initiated as a proxy conversion metric and also track the full funnel.
   
   **Your task:**
   1. Calculate metrics for each creative in ANALYSIS_DATA. Use the following formulas:
   - CTR (link click-through rate) = Link clicks / Impressions
   - CPC (cost per link click) = Amount spent (USD) / Link clicks
   - IPM (Installs per 1000 impressions) = App Installs / (Impressions / 1000)
   - Cost per App Install = Amount spent (USD) / App Installs
   - CR install to reg = Registrations completed / App Installs
   - Cost per registration completed = Amount spent (USD) / Registrations completed
   - Cost per add to cart = Amount spent (USD) / Adds to cart
   - Cost per checkout initiated = Amount spent (USD) / Checkouts initiated
   - If a denominator is zero, the resulting metric should be 0 or null.
   2. Analyze each creative from ANALYSIS_DATA according to the following rules:
   - Rule for determining statistical significance (significance):
   --Consider the data sufficient ("significance": "yes") if BOTH of the following conditions are met:
   --- Amount spent (USD) > 49
   --- Impressions > 10,000
   Otherwise, set "significance": "no".
   -Rule for evaluation (evaluation):
   -- Compare the key metrics of the test creative (CTR, CPC, Cost per App Install, Cost per checkout initiated) against the average values of the same metrics from the BENCHMARKS_DATA file.
   -- "high": The creative significantly outperforms the benchmark averages on key funnel metrics (e.g., higher CTR, and lower CPC and Cost per checkout initiated).
   -- "medium": The creative shows mixed results or performs on par with the benchmarks. For example, a good CTR but an expensive Cost per checkout initiated.
   -- "low": The creative significantly underperforms the benchmarks on most key metrics.
   3. Based on your analysis, provide a final JSON object containing the ad_id from the input data, and your calculated values for evaluation, significance, summary, and recommendation. For the summary, provide a 2-3 sentence explanation. If there is not enough data, set "significance" to "no" and explain why in the summary.

This template setup prompt: confirm you have configured the four credentials (facebookGraphApi, googleSheetsOAuth2Api, openAiApi, googlePalmApi), filled source, campaign_id (if applicable), and benchmarks_data CSV in the Set parameters node, then configure the three Google Sheets nodes to the same spreadsheet, set the Schedule Trigger, and activate.