Skip to main content

Follow-up

Follow-up covers the second turn (and beyond) within the same conversation: the user asks "why are you recommending the Fox?", "what's the difference vs the other one?", and the agent responds with a detailed explanation referencing specific products.

Types at.

How it works​

  1. The user asks a new question within the same conversationId.
  2. The backend identifies it as a follow-up (not a new search) and returns followUp: FollowUpResponse in SearchResponse.
  3. The widget renders with InlineFollowUp.

Shape​

interface FollowUpResponse {
  question: string                       // the user's question
  answer: string                         // the agent's explanation
  referencedProducts: FollowUpProduct[]
  recommendation: string                 // final verdict
}

interface FollowUpProduct {
  productId: string
  name: string
  price: number
  salePrice?: number
  currency?: string
  imageUrl: string
  url: string
  inStock?: boolean
  rating?: number
  reviewCount?: number
  brands?: string[]
  badge?: 'BESTSELLER' | 'TRENDING' | 'TOP_RATED'
  highlights: string[]                   // 2-3 highlights max
  verdict: string                        // contextual product verdict
  externalId?: string
}

.

  • Main block with formatted answer.
  • Collapsible list of referencedProducts, each showing highlights + verdict.
  • Final CTA "Show products" / "Hide products" (followUp.showProducts / hideProducts).
  • Footer with the highlighted recommendation.

Conversation state​

The widget keeps the conversationId received in the first response and forwards it on subsequent requests. If the user closes the widget and reopens it within the same browser session, the conversationId is preserved (sessionStorage). On the other hand, switching tab / SPA reload starts a fresh conversation.

i18n​

followUp.aiResponse, followUp.referencedProducts, followUp.recommendation, followUp.showProducts, followUp.hideProducts, followUp.viewProduct at

Last updated on