n8n with OpenAI API: Automate Content Generation at Scale

Problem: Generating Content at Scale Is Slow and Repetitive

You have 500 product descriptions to write. Or a weekly blog pipeline that pulls trending topics, drafts posts, and pushes them to your CMS. Doing this manually — or in a fragile Python script — doesn't scale.

n8n + the OpenAI API solves this. You get a visual workflow that triggers on a schedule or webhook, calls GPT-4o for generation, handles retries, and routes output to wherever you need it: Google Sheets, Notion, WordPress, Slack.

You'll learn:

• How to connect n8n to the OpenAI API with credential management
• How to build a reusable content generation workflow with dynamic prompts
• How to add error handling, retry logic, and output routing for production use

Time: 25 min | Difficulty: Intermediate

Why This Happens

n8n has a native OpenAI node since v1.22, but most tutorials stop at "send a prompt, get a reply." Production content workflows need more: dynamic prompts built from upstream data, rate limit handling, output parsing, and conditional routing based on the generated content.

What you're building:

• A trigger (schedule, webhook, or Google Sheets row)
• A prompt builder that injects dynamic variables
• An OpenAI node calling gpt-4o with a structured system prompt
• An output parser that extracts clean text
• A destination node (Sheets, Notion, WordPress REST API, or Slack)

Solution

Step 1: Install or Access n8n

If you're running n8n locally via Docker:

# Start n8n with persistent data volume
docker run -d \
  --name n8n \
  -p 5678:5678 \
  -v n8n_data:/home/node/.n8n \
  docker.n8nio/n8n:latest

Open http://localhost:5678 and create your account.

If you're on n8n Cloud, log in at app.n8n.cloud — skip Docker entirely.

Expected: n8n dashboard loads. You see the canvas editor.

Step 2: Add Your OpenAI Credential

1. In n8n, go to Settings → Credentials → New Credential
2. Search for OpenAI
3. Paste your API key from platform.openai.com/api-keys
4. Click Save — n8n encrypts and stores it

Credential name: OpenAI Production
API Key: sk-proj-...

If it fails:

• 401 Unauthorized → Your key is invalid or has no credits. Check platform.openai.com/usage.
• 403 Forbidden → Your key lacks the model:read scope. Generate a new key with full permissions.

Step 3: Create the Workflow Canvas

Click New Workflow. You'll build this node chain:

[Trigger] → [Set Variables] → [OpenAI] → [Parse Output] → [Destination]

Add a Schedule Trigger node first (or a Webhook node if you want on-demand generation):

• Schedule Trigger: Run every day at 8am UTC
• Webhook Trigger: POST to /webhook/content-gen with a JSON body

For this tutorial, use Manual Trigger while building — switch to Schedule or Webhook when ready for production.

Step 4: Build the Prompt with a Set Node

Add a Set node after the trigger. This is where you construct the dynamic prompt before passing it to OpenAI.

Configure two fields:

The || fallback means the workflow works even when fields aren't passed from upstream — useful for testing.

Step 5: Configure the OpenAI Node

Add an OpenAI node. Set:

• Credential: the key you saved in Step 2
• Resource: Chat
• Operation: Message a Model
• Model: gpt-4o (or gpt-4o-mini to cut costs by ~15x)
• System Message: ={{ $json.system_prompt }}
• User Message: ={{ $json.user_prompt }}
• Max Tokens: 600 (adjust for your word count targets)
• Temperature: 0.7 (lower = more consistent, higher = more creative)

{
  "model": "gpt-4o",
  "messages": [
    { "role": "system", "content": "{{ system_prompt }}" },
    { "role": "user",   "content": "{{ user_prompt }}" }
  ],
  "max_tokens": 600,
  "temperature": 0.7
}

Enable "Return Full Response" if you need token usage data for cost tracking.

If it fails:

• 429 Too Many Requests → You're hitting OpenAI's rate limit. Add a Wait node (2–5 seconds) before the OpenAI node, or enable the built-in retry in the node's settings (3 retries, 1s backoff).
• context_length_exceeded → Your prompt is too long. Trim the system_prompt or reduce word_count.

Step 6: Parse the Output

The OpenAI node returns a JSON object. The generated text lives at:

{{ $json.message.content }}

Add a Set node after OpenAI to extract and clean the output:

Now $json.generated_content holds your clean text for the next node.

Step 7: Route Output to a Destination

Connect a destination node. Three common options:

Option A — Google Sheets (log + review queue):

Add a Google Sheets node:

• Operation: Append Row
• Sheet: your content log sheet
• Columns: map topic, generated_content, tokens_used, generated_at

Option B — Notion (content database):

Add a Notion node:

• Operation: Create Page
• Database: your content pipeline DB
• Title: ={{ $json.topic }}
• Content block: ={{ $json.generated_content }}

Option C — HTTP Request to WordPress REST API:

POST https://yoursite.com/wp-json/wp/v2/posts
Authorization: Basic base64(username:app_password)

Body:
{
  "title": "{{ $json.topic }}",
  "content": "{{ $json.generated_content }}",
  "status": "draft"
}

Use n8n's HTTP Request node with the WordPress App Password (Settings → Users → Application Passwords).

Step 8: Add Error Handling

In production, OpenAI calls fail. Add a fallback path so your workflow doesn't silently drop items.

1. On the OpenAI node, click the ... menu → Add Error Output
2. Connect the error output to a Slack node (or Gmail node)
3. Configure it to send:

Content generation failed.
Topic: {{ $json.topic }}
Error: {{ $json.error.message }}
Workflow: n8n content pipeline

Now you get alerted on every failure instead of losing data.

Verification

Run the workflow manually with a test payload:

{
  "topic": "pgvector vs Qdrant for production RAG",
  "content_type": "blog introduction",
  "word_count": 150
}

You should see:

• The OpenAI node turns green with a response
• generated_content contains 130–170 words of clean text
• Your destination node (Sheets, Notion, etc.) shows the new row or page
• tokens_used shows a number between 200–400

Check token usage in your OpenAI dashboard to confirm API calls are registering:

platform.openai.com/usage

Scaling the Workflow for Batch Generation

To process a list of topics in one run, add a Google Sheets or Airtable node as the trigger instead of Manual/Schedule. Read all rows where status = "pending", then use a SplitInBatches node set to batch size 1 — this processes each topic sequentially and avoids parallel rate limit hits.

[Sheets: Read Rows] → [SplitInBatches (size=1)] → [Set Variables] → [OpenAI] → [Parse] → [Sheets: Update Row status="done"]

Add a Wait node (2 seconds) between SplitInBatches and OpenAI to stay under OpenAI's TPM (tokens per minute) limit on Tier 1 accounts.

What You Learned

• n8n's OpenAI node handles auth, retries, and response parsing — no SDK needed
• Dynamic prompts via Set nodes make one workflow reusable across content types
• Error outputs on the OpenAI node are essential for production — silent failures are worse than loud ones
• gpt-4o-mini at $0.15/1M input tokens is the right default for high-volume generation; switch to gpt-4o only for complex reasoning tasks

Limitation: n8n's OpenAI node doesn't support streaming responses. For real-time output to a UI, you'll need a direct HTTP Request node to api.openai.com/v1/chat/completions with "stream": true instead.

Tested on n8n 1.82, OpenAI API gpt-4o-2024-11-20, Docker on Ubuntu 24.04 and n8n Cloud