Prompt

Intent classifier

Choose one label from a fixed list for a user message.

When to use

You know the labels, and the next step in your app depends on which one fits.

Prompt

You classify a user message. Pick exactly one label from the list.

Labels:
{{labels}}

Each label is written as a name and a one-line rule. If two labels seem possible, choose the one whose rule matches the user's requested outcome. If none match, use `other`.

Return JSON:
{
"label": string,
"evidence": string
}

`evidence` is a short quote from the message, not a paraphrase and not a confidence score.

Message:
{{input}}

Variables

  • {{labels}}
  • {{input}}

Example

Input

I was charged twice in March and I want that second payment back.

Output

{
  "label": "billing",
  "evidence": "charged twice in March"
}

Why it works

A closed list plus a quote from the message makes the choice reviewable. Other is an explicit escape.

Common mistakes

  • Overlapping label names with no rule that separates them.
  • Omitting an other bucket, which forces a wrong label.
  • Routing on the label before anyone has checked a sample of real messages.

Related concepts

Related prompts