How to Fix Cannot Read Property 'choices' of Undefined in the OpenAI Node

The most common cause of an empty response is an invalid or expired API key, or a model name that does not exist. Go to the OpenAI node in your workflow, click on the credential, and verify the API key is correct. Then check the model field — if you are using a model like 'gpt-4' but your API account only has GPT-3.5 access, the API may return an error that n8n interprets as an empty response. Test the key by making a simple request via the OpenAI playground or a curl command.

Click the OpenAI node, go to Settings, and enable 'Continue On Fail'. Run the workflow again. Instead of crashing, the node will output the error object, which contains the actual API error message. Look at the output JSON for fields like error.message, error.code, or error.type. Common values include 'model_not_found', 'insufficient_quota', 'content_policy_violation', and 'server_error'. This raw error tells you exactly why the choices array was missing.

After the OpenAI node, add a Code node that validates the response structure before downstream nodes try to use it. This prevents the TypeError from propagating through your workflow. The Code node checks whether the expected fields exist and provides a fallback value if they do not. Set the mode to 'Run Once for Each Item'.

1const items = $input.all();
2const results = [];
3
4for (const item of items) {
5  const json = item.json;
6
7  // Check if this is an error from Continue On Fail
8  if (json.error) {
9    results.push({
10      json: {
11        text: '',
12        status: 'error',
13        error_message: json.error.message || 'Unknown OpenAI error',
14        error_code: json.error.code || 'unknown'
15      }
16    });
17    continue;
18  }
19
20  // Safely extract the response text
21  const text = json?.message?.content
22    || json?.text
23    || json?.choices?.[0]?.message?.content
24    || '';
25
26  results.push({
27    json: {
28      text: text,
29      status: text ? 'success' : 'empty_response',
30      model: json.model || 'unknown'
31    }
32  });
33}
34
35return results;

OpenAI's content moderation can block responses entirely, returning a response object with choices[0].finish_reason set to 'content_filter' and no message content. If your prompts might trigger content filters (medical, legal, or sensitive topics), you need to detect this case. Add logic in your Code node to check the finish_reason field and handle filtered responses gracefully by returning a user-friendly message instead of an empty string.

1const json = $json;
2
3// Check for content filter
4const finishReason = json?.choices?.[0]?.finish_reason || '';
5if (finishReason === 'content_filter') {
6  return [{
7    json: {
8      text: 'The AI model could not generate a response for this input due to content policy restrictions.',
9      status: 'content_filtered',
10      original_finish_reason: finishReason
11    }
12  }];
13}
14
15return [$input.item];

OpenAI API occasionally returns 500 or 503 errors during high load, which can cause empty response objects. Configure the OpenAI node's retry settings: click the node, go to Settings, and set 'Retry On Fail' to true with a max of 3 retries and a wait time of 2000ms between retries. This handles transient failures automatically without custom code. For more complex retry patterns, wrap the OpenAI node in a loop using the SplitInBatches node with a batch size of 1.