It feels like N8N is having some kind of issue. I tried using the text classifier and i just keep getting this error no matter what i did. what could be the problem? what do i do?
No idea. Please share your workflow and provide more context, @darlington_onuegbu
Essentially, your n8n workflow is crashing because of a âlanguage barrierâ between the AI and the Text Classifier node. The node is expecting a single, clean piece of data (an object), but the AI is delivering that data inside a list (an array). It is like asking someone for a single key, but they give you a locked box with the key inside; the node doesnât know how to open the box, so it simply gives up.
This happens because AI models often try to be âtoo helpfulâ with their formatting. Instead of just giving the raw answer, the AI is wrapping the result in Markdown code blocks (those triple backticks you see in the error) or square brackets. While this looks organized to a human reader, the n8n node is a strict machine that requires the data to be in one exact format to proceed.
The easiest way to fix this is to change the instructions you give the AI in the System Prompt. You need to be very firm and tell it to stop using any fancy formatting. Specifically, tell it to provide âraw JSON onlyâ and explicitly forbid it from using markdown code blocks or wrapping the answer in a list. This usually forces the AI to strip away the extra characters that are causing the error.
In your System Prompt Template, add the following instruction at the end:
"Return ONLY a raw JSON object. Do NOT wrap the response in markdown code blocks (like ```json). Do NOT wrap the object in an array. Output only the JSON."
Beyond the prompt, there is a technical setting in your node called âUse Responses APIâ that is currently turned on. This setting changes the way n8n and OpenAI communicate. Turning this off may change the structure of the answer the AI sends back, which could potentially resolve the conflict and allow the Text Classifier to read the data correctly.
If neither of those steps work, the Text Classifier node might simply be too rigid for the model you are using. In that case, the best backup plan is to use an âAI Agentâ node instead. Agents are much more flexible with how they handle information. You can then follow the Agent with a simple âCodeâ node to manually clean up the AIâs response, ensuring your workflow stays running regardless of how the AI formats its answer.
Before assuming it is n8n itself, the Text Classifier node erroring âno matter whatâ is usually the model call underneath it failing, not the node. The classifier sends your text to the connected LLM and parses the result, so the error is often the credential, the model name, or the response shape.
Quick things to check: open the error detail (click the failed node, look at the full message, not just the red banner) and see if it mentions auth, rate limit, or parsing. Confirm the connected model credential still works by testing it in a plain LLM node on its own. And make sure your categories are defined, an empty or malformed category list makes the classifier throw.
Post the actual error text and your node setup and it is usually obvious. âSome kind of issueâ could be five different things, the exact message narrows it fast.