Answer Feedback

When to useCopy Anchor Link

• You finished a substantive answer (research, a plan, a recommendation) and want to know how it landed.
• The host keeps ratings, or you'll use them to improve the next answer in the conversation.

When not to useCopy Anchor Link

• The reply is a line or a confirmation. A rating adds noise.
• You're waiting for a decision. Show the Approval Request or a question instead.
• It would appear under every message. Offer it once per substantive answer.

How it worksCopy Anchor Link

Two icon buttons (and a third for Regenerate, when you set regenerate). A thumbs up sends at once. A thumbs down opens a short list of reasons as chips, and Send sends the ones the user ticked, or none. After a rating the row is replaced by a one-line record, like "Thanks. Too long, Out of date.", and focus moves to it.

The reasons are a closed list, so the next turn gets something typed to act on instead of a bare "unhelpful". The defaults are Not accurate, Wrong source, Too long and Out of date; set reasons to what could actually be wrong with this answer (for a sourced answer: wrong source, out of date).

How the agent should use itCopy Anchor Link

• Name the likely failures in reasons, 2–6 of them, a few words each.
• On a down rating, fix what was named (shorten, re-check sources, refresh) and don't ask why again.
• On regenerate, answer the question afresh, differently from before. Set regenerate only when a second try is cheap.
• Echo answerId (a message id) so a stored rating points at its answer.

InteractionCopy Anchor Link

Each result also carries answerId when you set one.

InstallationCopy Anchor Link

npx shadcn@latest add https://your-site.com/r/answer-feedback.json

UsageCopy Anchor Link

import { AnswerFeedback } from "@/components/genui/answer-feedback"

export const catalog = {
  rate_answer: AnswerFeedback,
} satisfies GenUICatalog

import { tool } from "ai"
import { answerFeedbackTool } from "@/components/genui/answer-feedback.schema"

const tools = {
  // No execute(): the call is completed by the user's rating in the UI.
  [answerFeedbackTool.name]: tool({
    description: answerFeedbackTool.description,
    inputSchema: answerFeedbackTool.inputSchema,
  }),
}

SchemaCopy Anchor Link

Host props: status, onResult, result and className.

SourceCopy Anchor Link