How to send user metadata along with prompts to Gemini from n8n?

You send user metadata to Gemini from n8n by adding it directly into the JSON you send in the HTTP Request node that calls the Gemini API. Gemini supports arbitrary metadata via the generationConfig or system\_instruction fields, or by just embedding it as part of the contents data. n8n does not restrict what JSON you send, so you simply include metadata in the request body along with the prompt.

 

What this looks like in practice

 

In real n8n workflows, you typically:

• Collect the user metadata (from a webhook, database query, form submit, etc.).
• Merge it with the prompt in a Set node or by using expressions inside the HTTP Request node.
• Send everything together as JSON to Gemini’s Generative Language API.

Gemini accepts extra fields without complaint as long as the required fields (contents) are valid. This is the easiest and most stable way to attach metadata for downstream logging, auditing, context control, or user‑specific behaviors.

 

A clean, safe way to send metadata

 

Below is a fully valid request body you can use inside an n8n HTTP Request node (POST → https://generativelanguage.googleapis.com/v1beta/models/gemini-pro:generateContent). It embeds metadata in a custom field userMetadata, and also shows the normal Gemini contents block.

 

{
  "userMetadata": {
    "userId": "{{$json.userId}}",
    "plan": "{{$json.plan}}",
    "language": "{{$json.language}}"
  },
  "contents": [
    {
      "role": "user",
      "parts": [
        {
          "text": "{{$json.prompt}}"
        }
      ]
    }
  ],
  "generationConfig": {
    "temperature": 0.7
  }
}

 

This works because Gemini’s API does not reject unknown top-level fields. n8n will replace the expressions like {{$json.userId}} with actual data from previous nodes.

 

Alternative: Put metadata inside the system instruction

 

If you want Gemini to actually use the metadata in reasoning (for personalization, policy, etc.), put it in system\_instruction. This is fully supported and production-safe:

 

{
  "system_instruction": {
    "role": "system",
    "parts": [
      {
        "text": "User metadata:\nUser ID: {{$json.userId}}\nPlan: {{$json.plan}}\nLanguage: {{$json.language}}"
      }
    ]
  },
  "contents": [
    {
      "role": "user",
      "parts": [
        {
          "text": "{{$json.prompt}}"
        }
      ]
    }
  ]
}

 

This makes the metadata visible to the LLM itself, which is sometimes necessary (e.g., enforcing per‑plan rules or personalizing tone).

 

How to prepare the metadata cleanly in n8n

 

• Use a Set node to structure the user data cleanly. This prevents messy nested JSON or broken expressions.
• If metadata comes from a webhook, check its format using an If node before sending to Gemini so you don’t break the HTTP request.
• If metadata is optional, wrap the expression with {{$json.fieldName || ""}} so the JSON body never fails.

 

Important production notes

 

• Gemini’s API does not enforce strict schemas, so adding metadata fields is safe.
• Be mindful of personally identifiable information (PII). If sending real user IDs, ensure secure storage of credentials and HTTPS is enabled.
• If using n8n Cloud or Webhook trigger, sanitize incoming inputs before injecting them into prompts or metadata.

The core idea is simple: you send metadata by including it in the JSON body of your Gemini API call from the HTTP Request node. n8n lets you freely structure the JSON, and Gemini gracefully accepts and forwards any extra fields.