Conversational commerce

Vertex AI Search for commerce offers two conversational capabilities: conversational product filtering and the conversational commerce agent. Conversational product Filtering guides users with structured questions to refine searches, while the conversational commerce agent enables open-ended chat experiences.

Two conversational capabilities with a single API endpoint

Both capabilities are powered by the conversationalSearch API endpoint.

Conversational product filtering

  • Use case: Guides users to refine broad searches, such as dress, with clarifying questions like What is the color?

  • Interaction model: Reflects structured, multiple-choice questions and answers.

Conversational commerce agent

  • Use case: Enables open-ended, natural language chat experiences like Help me plan a party.

  • Interaction model: Includes AI-generated text responses, dynamic search suggestions, and follow-up questions.

Enable conversations with parallel API calls

The Conversational API does not return product search results. Its purpose is to provide the conversational experience.

Your application must make two separate, parallel API calls for each user interaction. This architecture is fundamental to both product filtering and the conversational commerce agent.

  • Conversational API

    • Endpoint: Is conversationalSearch.
    • Purpose: Enables follow-up questions, text responses, and refined queries.
    • Returns: Fields include conversation_id, followup_question, and conversational_text_response.

  • Search API

    • Endpoint: Is search.
    • Purpose: Gets the list of products to display to the user.
    • Returns: Expect a standard list of product results.

Uses the core field conversation_id

The conversation_id is the key to maintaining the state and history of a single conversation session.

  • First request: Sends the initial query with no conversation_id.
  • Response: The API returns a new conversation_id.
  • Your action: Store this in client-side session storage.
  • Subsequent requests: For all follow-up interactions within the same session, you must send the stored conversation_id in the request. This allows the system to understand context (such as, the user already said they want a yellow dress).

A new conversation is started with just a request without a conversation_id.

What's next