How to Connect OpenAI API with n8n to Build Custom AI Workflows

Standard rule-based automation is great for moving data from point A to point B. But when your workflows require decision-making, text generation, or natural language understanding, traditional triggers and actions hit a wall.

By connecting the OpenAI API with n8n, you transform simple data pipelines into intelligent, autonomous workflows. Building on our previous guides like 7 n8n Workflows Every Small Business Should Steal and How to Automate Sales Tracking with Google Sheets and n8n, this step-by-step tutorial will show you how to set up OpenAI API credentials in n8n, configure custom calls, and build production-ready AI automations.

Why Connect OpenAI API Directly to n8n?

While n8n features pre-built LangChain and AI nodes, making direct API calls or using the dedicated OpenAI node gives you total control over:

• Model Selection & Cost Tuning: Switch seamlessly between flagship and lightweight models to optimize cost and latency.
• Structured Outputs: Force OpenAI to return valid JSON schema that subsequent n8n nodes (like Google Sheets, Airtable, or Slack) can parse effortlessly.
• Custom System Prompts: Maintain strict control over context, guidelines, and output formatting.

Prerequisites

Before setting up the workflow, ensure you have:

• An active OpenAI account with API credit balance and a generated API Key (found in your OpenAI Developer Dashboard).
• An operational n8n instance (either n8n Cloud or self-hosted).

Step 1: Add OpenAI Credentials in n8n

To interact with OpenAI safely without hardcoding your API key into workflow nodes:

1. Open your n8n dashboard and navigate to Credentials in the left sidebar.
2. Click Create Credential and search for OpenAI API.
3. Paste your secret key starting with sk-... into the API Key field.
4. Save the credential as OpenAI Production Account.

Step 2: Configure the OpenAI Node (or HTTP Request Node)

You can use either the native OpenAI Node or a generic HTTP Request Node. Here is how to configure the native node for maximum reliability:

1. Add an OpenAI Node to your workflow canvas.
2. Set the Resource to Chat and Operation to Complete.
3. Select your saved OpenAI Credential.
4. Choose the Model — a lightweight/mini-tier model for fast, simple processing, or the current flagship model for complex reasoning. OpenAI’s lineup has moved past the GPT-4o generation, so check the model list in the node itself for what’s current.
5. In the Messages section, define two roles:
   • System Prompt: Sets the persona and operational boundaries.
   • User Prompt: Dynamically pulls input from previous nodes using n8n expressions (e.g., {{ $json.email_body }}).

Step 3: Enforce JSON Output Parsing

To prevent raw conversational text from breaking downstream applications, instruct OpenAI to return structured JSON.

System Prompt Example:

You are an AI assistant analyzing customer inquiry emails. 
Analyze the input text and respond ONLY in valid JSON matching this schema:
{
  "sentiment": "positive | neutral | negative",
  "urgency_score": 1-5,
  "category": "billing | technical_support | sales | general",
  "summary": "1 sentence summary"
}
Do not include markdown formatting or extra text outside the JSON object.

In n8n, add a JSON Parser Node or use a Code Node (JSON.parse($json.message.content)) right after the OpenAI node to transform the LLM output into native n8n data fields.

Practical Use Case: Automated Urgent Lead & Support Routing

Here is a real-world workflow architecture combining n8n and OpenAI:

[ Webhook / Email Trigger ] 
            │
            ▼
[ OpenAI Node (Analyze Sentiment & Category) ]
            │
            ▼
[ Code Node (Parse JSON & Extract Variables) ]
            │
            ▼
[ Switch Node (If Urgency >= 4) ]
     ├── YES ──► [ Send Immediate Slack Alert ]
     └── NO  ──► [ Log Inquiry to Google Sheets ]

How It Works:

• Trigger: A new email or contact form submission triggers the workflow.
• AI Processing: OpenAI evaluates the sentiment, urgency level, and main category of the message.
• Branching Logic: If urgency_score is 4 or 5, n8n triggers an immediate high-priority alert to your team on Slack. Otherwise, it logs the inquiry neatly into Google Sheets for standard processing.

Best Practices & Cost Optimization

• Use a lightweight model for classification tasks: For simple categorization, extraction, or sentiment analysis, the smaller “mini” or “nano” tier in OpenAI’s current lineup provides nearly identical performance at a fraction of the cost of the flagship model — check OpenAI’s current pricing page before locking in a model for production.
• Set temperature control: Set temperature to 0.0 or 0.2 when you need consistent, deterministic outputs (like JSON parsing). Use higher temperatures (0.7+) only for creative drafting.
• Implement error handling: Attach an Error Trigger Node in n8n. If OpenAI hits a rate limit (HTTP 429) or context window constraint, n8n can automatically retry or alert you without failing silently.