How to debug why my OpenAI response is empty in n8n?

If your OpenAI node in n8n is returning an empty response, the fastest way to debug it is to open the execution, look at the Incoming Data and the Output of the OpenAI node, and confirm whether the API actually returned a normal JSON body or if n8n received an empty string, an error hidden in json.error, or the model's output is located under a different field than the one you’re reading. Most empty responses come from sending the wrong input format, asking for a field that doesn’t exist, or hitting a silent error (model not allowed, token limit, or missing "messages" array).

Step‑by‑step ways to debug an empty OpenAI response in n8n

Below are the most reliable checks I use in production when an OpenAI node suddenly comes back empty. These steps work for both the OpenAI (Chat) node and the older OpenAI or HTTP Request setups.

• Check the execution's OpenAI node output directly. Open the node → click the tab “Output” → see if json contains anything like choices. If the output panel is literally empty, the API never returned usable data.
• Look for hidden errors in the same node. Sometimes n8n catches the error inside json.error instead of failing the execution. Expand all fields; common values are invalid_request_error, context_length_exceeded, or model_not_found.
• Verify you’re selecting the correct field in Expressions. Most empty results are actually because someone tries to read something like {{$json\["choices"]\[0]["text"]}} but the real structure is chat.completions style, for example: \`\`\`json { "choices": [ { "message": { "content": "Hello!" } } ] } \`\`\` If you call $json.choices[0].text, that field does not exist — n8n will show it as empty.
• Confirm that your input to OpenAI is valid JSON. Chat models need a messages array. If you accidentally send a single string or an empty array, OpenAI can return an empty output or an error. A valid body looks like: \`\`\`json { "messages": [ { "role": "user", "content": "Hello" } ] } \`\`\`
• Check your model choice. Some models are not enabled in your account or require upgraded billing. If the model is wrong, n8n may show a silent error payload.
• Enable “Always Output Data” for debugging. In many n8n AI nodes, turning this option on makes the node output the raw response even if it thinks it is empty. That helps you see exactly what the API returned.
• If using HTTP Request instead of the built‑in node, inspect the raw response in binary. Occasionally OpenAI responds with a streamed format or includes whitespace. Click “Binary” tab and inspect any attached data.
• Make sure you’re not accidentally working with multiple items. If a previous node outputs multiple items, the OpenAI node will run once per item. The “empty” one might just be one of the items while another contains the real data.
• Check timeout or rate limit errors. In logs you may see 429 or a long execution time. n8n sometimes converts these into empty objects depending on configuration.
• Run the OpenAI call inside a test Function node. This lets you see exactly what the model returns. Example: \`\`\`javascript // This code runs inside an n8n Function node return [ { json: { sampleInput: $json } } ]; \`\`\` This is not calling OpenAI, but it helps verify the JSON going into OpenAI is what you expect. Many “empty output” issues are caused by wrong or missing input.

The core idea

An “empty response” in n8n is almost never the OpenAI model being blank; it's almost always a mismatch between what OpenAI returns, what n8n expects, or what your expression is reading. You fix it by checking the node’s output JSON, validating the structure, confirming the input messages, and ensuring your expression matches the real path.