Hey everyone I have questions about text classifier
I’m stuck keep sending to one node gmail I have 4 gmail node in 4 categories when I am doing execute (send me to last node how I am suppose change to first node?)
do someone have suggestions for me
Hey everyone I have questions about text classifier
I’m stuck keep sending to one node gmail I have 4 gmail node in 4 categories when I am doing execute (send me to last node how I am suppose change to first node?)
do someone have suggestions for me
Hi @Donovan2212 , I think the first thing I would check is whether each Gmail node is connected to the correct output/category from the Text Classifier.
I would not try to “change to the first node” manually during execution. Instead, I would make sure the first Gmail node is connected to the first classifier output, the second Gmail node to the second classifier output, and so on. Then I would test with very clear sample texts for each category and check the Text Classifier output to see which category is actually being returned.
If it always goes to the last Gmail node, it may also mean that the last output is acting like your fallback or default path, or that the category descriptions are not different enough. Try renaming the categories with very explicit labels and add examples in the descriptions, such as “Use this category only when the email is about invoices or payments”.
If you can share a screenshot of the Text Classifier categories and how the four Gmail nodes are connected, it will be easier to see whether this is a routing issue or a classification prompt issue.
Hey @tamy.santos
There is my screenshots well im trying to do cursus learning and it suppose look like this but executed wrong node.
Hey @Donovan2212 , thanks for the screenshot.
From what I can see, the workflow is not randomly choosing the wrong Gmail node. The Text Classifier is classifying your input as Others, because the green execution path and the 1 item are going through the Others output.
I would check three things.
First, make sure the Gmail nodes are active. In your screenshot they show as deactivated, so even if the classifier sends the item there, the Gmail node may not actually run.
Second, open the Text Classifier node and make the category descriptions very explicit. For example, do not only use category names like Payments or Service Request. Add clear descriptions such as “Use this category when the email is about invoices, payments, refunds, billing, or transaction issues”.
Third, test with very obvious text first. For example, send something like “I need help with my payment invoice” and check if it goes to Payments. If it still goes to Others, the model is not getting enough context from the categories.
As a workaround, if this is for a course or a simple learning workflow, I would temporarily use an IF or Switch node with keyword rules instead of the Text Classifier. For example, if the email contains “payment”, “invoice”, or “billing”, send it to Payments. Once that works, you can replace it with the Text Classifier again and improve the category descriptions.
So in short, the node is going to Others because that is what the classifier selected. I would activate the Gmail nodes, improve the classifier category descriptions, and test each category with very clear sample emails.
Hey @Donovan2212, solid workflow idea - classifying emails and routing them automatically is one of my favorite n8n use cases!
On top of what @tamy.santos covered, there’s a quick debugging trick that really helps: add a Set node right after the Text Classifier (before your Gmail nodes) and output the classifier’s category result as a field. This lets you run the workflow manually on a test email and see exactly what category was assigned before anything gets sent.
Something like:
{{ $json.category }}
or
{{ $json.output }}
(the field name depends on your n8n version - just check what the Text Classifier outputs in the execution view)
Also one more note: the Text Classifier by default uses the LLM you configured (usually GPT). If your category descriptions are too generic or similar to each other, the model may default to the last one as a “catch-all”. Try making each category description as distinct as possible and add a few example emails in the description field.
Can you share a screenshot of your Text Classifier node settings? Seeing the category names and descriptions would help narrow down the routing issue.
deactivated was doing testing then setups parameters red triangle error i know I wasn’t done there trying first fixing with the execute text classifier to gmail node it should to category 1 not 4
@nguyenthieutoan @tamy.santos
This is what my parameters wrote and my others 3 gmails already activated turning on (executed showed me to last execution view node gmail 4)
You should not try to manually force it to go to Gmail 1. The Text Classifier needs to understand category 1 more clearly, and Others should be treated as a real fallback only when the email does not match the other categories.
I would make the category 1 description more explicit, for example: “Use this category when the email is asking about consulting services, available services, pricing, rates, service details, or availability.” And I would make Others stricter, for example: “Use this only when the email does not match service requests, consultation scheduling, or payment questions.”
@Donovan2212 Good progress getting the workflow set up! Looking at your parameters screenshot, I can see exactly what’s happening.
The issue is that your category descriptions are too short and generic, so the LLM defaults to “Others” when it’s uncertain. The Text Classifier works by semantic matching, not keyword matching, so the description quality really matters.
Try updating like this:
Category 1 - Services Request:
“Use this category when the email is asking about any service you offer, including pricing, rates, availability, what services are included, or how your consulting works.”
Category 2 - Consultation Request:
“Use this category when the email wants to book, schedule, or arrange a meeting, call, or initial consultation with you.”
Category 3 - Payments:
“Use this category when the email is about invoices, payment status, refunds, billing, or any financial transaction.”
Category 4 - Others:
“Use this ONLY as a last resort for emails that clearly do not fit any of the above three categories.”
Also, that red triangle error before was likely from an unconfigured node. Make sure all nodes have valid credentials and required fields filled in before testing. Once you clean up the descriptions, run it again with a test email like “How much do your consulting services cost?” and it should go to Category 1.