How to Debug Why Your OpenAI Response Is Empty in n8n

Open the execution that produced the empty response by going to Executions in the left sidebar and clicking on the relevant run. Click on the OpenAI node in the execution view to see its input and output data. Look at the output panel on the right side. If the output shows an empty JSON object or an empty string in the message.content field, the API did return a response but it was empty. If you see no output data at all, the node may have errored silently. Check the raw response by switching to the JSON view (click the {} icon in the output panel). Look specifically for the choices array: if it is empty or the first choice has an empty content string, the issue is on the OpenAI side.

Replace the dynamic expression in your OpenAI node's prompt field with a simple hardcoded string like 'Say hello world'. Run the workflow manually. If this returns a valid response, the problem is with your dynamic input, not the OpenAI configuration. If even the hardcoded prompt returns empty, the issue is with your API key, model selection, or account status. This test eliminates the entire upstream workflow from the debugging process and focuses on the OpenAI node itself. After confirming the hardcoded prompt works, gradually reintroduce dynamic elements to identify which expression causes the empty response.

In the node that reads the OpenAI response, check the expression you are using. Common mistakes include referencing $json.text when the actual field is $json.message.content, or using $('OpenAI').first().json.text when the output structure uses a different key. Click on the downstream node, open the expression editor, and use the variable selector on the left to browse the actual output structure of the OpenAI node. The correct path for the OpenAI Chat Model sub-node output is typically $json.output or $json.text depending on how it is connected. For the standalone OpenAI node, it is usually $json.message.content.

1// Common correct expressions for OpenAI output:
2
3// If using OpenAI Chat Model sub-node with AI Agent:
4{{ $json.output }}
5
6// If using standalone OpenAI node (HTTP Request):
7{{ $json.choices[0].message.content }}
8
9// If referencing from a non-adjacent node:
10{{ $('OpenAI Chat Model').first().json.output }}
11
12// WRONG (common mistakes):
13// {{ $json.text }}  — field doesn't exist
14// {{ $json.response }}  — field doesn't exist
15// {{ $json.data.choices[0] }}  — wrong nesting

OpenAI's content moderation can silently return empty responses when the prompt or expected output triggers safety filters. In the raw API response, look for a finish_reason field. If it says 'content_filter' instead of 'stop', OpenAI blocked the response. Also check for a content_filter_results object in the response. To fix this, review your system prompt and user prompt for content that might trigger filters. Rephrase the prompt to avoid flagged topics. If your use case is legitimate, you can apply for modified content policy access through OpenAI's platform. Add a Code node after the OpenAI node that checks the finish_reason and routes filtered responses to an alternative handling branch.

1// Code node: Check for content filter
2const items = $input.all();
3const results = [];
4
5for (const item of items) {
6  const finishReason = item.json.finish_reason || item.json.finishReason || '';
7  const content = item.json.message?.content || item.json.output || '';
8
9  results.push({
10    json: {
11      ...item.json,
12      wasFiltered: finishReason === 'content_filter',
13      wasEmpty: content.trim() === '',
14      finishReason: finishReason
15    }
16  });
17}
18
19return results;

Insert a Code node between your data source and the OpenAI node that logs the exact input being sent. This lets you see the complete prompt including all dynamic expressions after they have been resolved. Set the Code node to Run Once for All Items. Log the input data and then pass it through unchanged. After running the workflow, check this node's output in the execution data to see exactly what the OpenAI node received. Often you will find that a variable resolved to undefined or null, resulting in a prompt like 'Summarize the following: undefined' which may cause the model to return empty content.

1// Code node: Diagnostic Logger
2// Place between data source and OpenAI node
3const items = $input.all();
4
5for (const item of items) {
6  console.log('OpenAI input payload:', JSON.stringify(item.json, null, 2));
7
8  // Flag suspicious values
9  const prompt = item.json.prompt || item.json.text || item.json.message || '';
10  if (!prompt || prompt.includes('undefined') || prompt.includes('null')) {
11    console.log('WARNING: Prompt contains empty or undefined values');
12  }
13}
14
15// Pass through unchanged
16return items;

An expired, revoked, or quota-exceeded API key can sometimes produce empty responses instead of clear error messages, especially when using older n8n node versions. Go to the OpenAI platform dashboard at platform.openai.com and check your API key status, billing, and usage limits. Verify that the key has access to the model you selected in the n8n node. If you recently rotated keys, update the credential in n8n by going to Credentials in the left sidebar, finding your OpenAI credential, and entering the new key. Test the updated credential by running the workflow with the hardcoded prompt from step 2.