I’ve been using the Text Classifier node (v1.1) quite heavily in production workflows, and there are a couple of things I think would make it significantly more reliable and powerful.
1. Fallback AI model support
Right now, the Text Classifier node does not have any fallback mechanism when the AI call fails. Unlike other AI nodes in n8n that have a built-in fallback option, the classifier just throws an error and breaks the entire workflow.
This is a real problem when you run batch classification at scale. For example, if you’re classifying 500+ messages in a loop and one API call times out or hits a rate limit, the whole run fails. There’s no way to catch that gracefully at the node level.
Other AI nodes in n8n support a fallback model connection, so if the primary model fails, it retries with a secondary one. The Text Classifier node should have the same behavior.
2. Additional classification mechanism ideas
Since this node is often used in critical decision paths (routing customer messages, categorizing support tickets, filtering leads, etc.), it would be great to also consider:
- Confidence score output: Let the node output a confidence value alongside the category label, so downstream logic can handle low-confidence cases differently (e.g., escalate to human)
- Custom instructions field: A way to pass additional context or persona to the classification prompt, similar to how the AI Agent node lets you customize the system prompt
Why this matters
The Text Classifier is one of the most useful nodes for building smart workflows, but right now it feels fragile compared to other AI nodes. Adding fallback support would bring it in line with the rest of the AI node family and make it production-safe.
Would love to hear if others have run into the same issue, and if the team has any plans in this direction. Happy to provide more context or examples from my production setup.