Score customer support AI responses with GPT‑4 judge metrics

Score open-ended AI responses with a judge model. This template shows how to evaluate a customer support agent using a separate LLM that rates each response on correctness and helpfulness, going beyond what exact match scoring can capture.

What you'll do

Open the workflow and review the production path (chat trigger, AI Agent generates a support response, response returned to the user). Open the Evaluations tab and click Run Test to feed question + expected answer pairs through the AI Agent. Watch the judge model score each response on correctness (1-5) and helpfulness (1-5). Review per-test-case scores in the Evaluations tab alongside token usage and execution time.

What you'll learn

How LLM-as-a-Judge works and when it beats deterministic scoring How to wire a separate judge model into your evaluation path How to write a custom scoring prompt that returns a numeric score and a justification When to use n8n's built-in Correctness and Helpfulness metrics versus a custom judge

Why it matters

Customer-facing responses are subjective. A response can be technically accurate but tonally wrong, or polite but useless. LLM-as-a-Judge gives you a measurable signal for the kind of quality that matters but resists simple matching, so you can iterate on prompts with confidence instead of guesswork.

This template is a learning companion to the Production AI Playbook, a series that explores strategies, shares best practices, and provides practical examples for building reliable AI systems in n8n.