Conversational commerce

Preview: This feature is in Private Preview and is scheduled to be generally available at a later date.

This page describes a guided search feature in Vertex AI Search for commerce conversational commerce.

Conversational commerce uses conversational product filtering to provide users with a real-time, ongoing conversational experience, enabling a more interactive search experience.

The conversational commerce feature functions as part of the guided search package, benefiting customers by engaging them in a conversation to narrow down search queries faster by presenting them with relevant products or follow-up questions.

What is conversational commerce?

Conversational commerce is an AI-driven guided search and product discovery. Instead of searching with keywords, users utilize natural language to ask for what they need—complete with follow up questions, multimodal interactions, improved intent understanding, and grounding with data beyond the product catalog. This approach enables more intuitive and efficient results filtering, helping users get to exactly what they are looking for.

Impact of conversational search

Guided search is a Vertex AI Search for commerce package that helps you create a more intuitive, efficient, and conversion-oriented shopping experience for your site visitors. As a central part of the guided search package, conversational commerce improves search result relevance and reduces user friction by engaging shoppers in conversations.

Natural language understanding and multimodal inputs enable site visitors to discover relevant products faster. Personalized experiences, contextual awareness, and real-time results further engage users, increasing the likelihood of purchase completion. These improvements create an integrated user journey, ultimately leading to increased customer loyalty, reduced bounce rates, and higher merchant site conversions.

The role of external search and AI assistants

When a user has a preferred retailer, they tend to go directly to it. However, when exploring new or broader options, the customer user journey is more likely to first start at a marketplace orientation, such as Google Search or an AI assistant.

As AI-driven search evolves, the Vertex AI Search for commerce guided search package bridges the gap between external discovery, whether through legacy Search or AI assistants, retailer sites or apps. By treating the transition as a single, fluid conversation, conversational commerce ensures users find relevant results quickly, leading to increased conversion rates.

Conversational commerce features

Conversational commerce adds to the Vertex AI Search for commerce experience in the following ways:

• Narrows user queries faster: Conversational commerce filters 10,000 products down to less than 100 products, increasing the likelihood that the user decides to make a purchase.
• Hyper-personalization: Search agents analyze shoppers' preferences, purchase history, and social media activity to provide more personalized product recommendations, promotions, and shopping experiences.
• Integrated end-to-end journeys: From product discovery to checkout, the search agents accompany the end user along their entire shopping journey with immersive, dynamic, and continuous conversation.
• Adapted to commerce use cases: Conversational commerce covers ecommerce, product discovery, and support use cases.
• More immersive user experiences: With the help of search agents threading user conversations in the background, augmented and virtual reality can be additionally implemented on the merchant site to create shopper experiences such as virtual try-ons, store tours, and spatial product visualization for users.

Mobile-first experience

Nearly 80% of all ecommerce visits worldwide occurred on a mobile phone in 2024. Smaller screens, shorter user sessions and clustered menus create unique challenges for legacy search experiences. Conversational commerce ad is designed to enable users to access the full power of AI-driven search from their mobile devices.

Multimodal search inputs

Conversational commerce enables users to search using multimodal input methods such as voice and image in addition to text. It understandings user intent, context and natural language variations in phrasing. Ensure seamless integration between input methods without losing context.

Voice search

Spoken queries are often structured differently than typed ones. Vertex AI Search for commerce processes these variations while accounting for variables like accents, background noise and "um's," "uh's" and "like's." For mobile, voice search is not only easier to input, but it can also take up less screen space, allowing more real estate product visuals.

Image search

Image recognition makes it faster to find a similar or unique item on a social media post or snapping a picture in real life. Shoppers can use image search to quickly search and locate a similar item on your site.

Core principles and best practices

This section describes core principles and best practices for using conversational commerce as part of your guided search package.

Clarity and transparency

Configure your site in such a way that the shopper understands why results appear and has the ability to refine or adjust search parameters.

• Ambiguous queries: When encountering a vague search, the conversational search proactively seeks clarification.
• Nuanced queries: Extremely detailed queries require strong content structures and detailed metadata for Vertex AI Search for commerce to return accurate results.
• User expectations: How and why specific results were prioritized must be shown with transparency for results based on the site visitor's search history.
• Transparent limitations: If a query can't be understood or if it returns only limited results, Vertex AI Search for commerce lets the site visitor know and offers constructive alternatives.
• Efficiency and speed: Delivers a seamless search experience, particularly in ecommerce.
• Predictive search: Provide autocomplete suggestions, predict intent, and surface relevant results instantly using historical data, trending queries, and user behavior.
• Enhance filtering mechanisms: Refine results through conversational inputs to accelerate the process of getting what you are looking for.
• Intelligent ranking and optimization: Prioritize the most relevant products based on context, user history, and search trends.
• Reduced cognitive load: Configure Vertex AI Search for commerce to harness contextual awareness to remember a user's previous selections to minimize redundant actions and accelerate user decision-making.

Use dynamic filtering to create personalization without overload

To avoid overwhelming users, dynamic filtering can strike a balance between personalization and user autonomy. Instead of overwhelming a user with too many filters upfront, conversational commerce has the capability of suggesting refinement based on preferences from a user's previous search or purchase data. Additionally, Vertex AI Search for commerce is adaptable to real-time behavior. For example, purchase history may not be relevant while searching for a gift for someone else, so conversational commerce can recognize the contextual changes and adjust accordingly.

Control features to ensure context and continuity across sessions

Rather than treating every query in isolation, contextual awareness allows for more efficient searches without repeating previous inputs. Not only should contextual awareness be pervasive within and across sessions, but also across devices to ensure continuity.

• Continuity within sessions: Vertex AI Search for commerce is designed to remember a user's past interactions within a session to allow for incremental refinements.
• Continuity across sessions: Site visitors should be able to pick up where they left off in a previous session without conversational commerce feeling intrusive.

Integrate accessibility and inclusivity considerations

Whether site visitors interact through voice, text, or images, Vertex AI Search for commerce must provide inclusive solutions that support individuals with varying abilities, preferences, or technological constraints.

Accessibility tools

Multimodal input support	Voice-to-text correction
Screen reader compatibility	Real-time transcription
Clear semantic structuring	Predictive text assist
Voiceovers for image-based content	Autocorrect

Use targeted questions to ensure graceful error handling

Inevitably, there are instances when a query produces no results on a particular site. Instead of an unhelpful no results message, conversational commerce offers intelligent suggestions and alternatives. Additionally, an conversational commerce prompts for clarification, asking targeted questions to narrow down preferences and needs, creating a user-centric dialogue.

Design calls-to-action to end conversations effectively

To prevent user abandonment, conversational commerce ends conversations with clear, actionable pathways, maximizing user satisfaction and boosting the likelihood of conversion by maintaining momentum. Furthermore, conversational commerce can create opportunities for re-engagement after the initial interaction by surfacing relevant follow-up questions based on browsing activity or purchase history.

Conversational commerce APIs

Conversational commerce is supported only by the Conversational API. The conversationalFilteringMode in the Conversational API distinguishes between conversational commerce and conversational product filtering.

• gRPC: conversationalSearchService
• REST: conversationalSearch

User journeys and query classifications

Conversational commerce uses search query categories to determine whether or not an LLM-based answer is generated and how user queries are handled by the Conversational and Search APIs for these user scenarios:

View table

User journeys	Classifications and examples	Handling
Irrelevant queries that don't require an LLM answer Sample Conversational API request placement: "projects/118220807021/locations/global/catalogs/default_catalog/placements/default_search" branch: "projects/118220807021/locations/global/catalogs/default_catalog/branches/default_branch" query: "Are there any deals for Thanksgiving?" conversational_filtering_spec { conversational_filtering_mode: ENABLED } visitor_id: "test" Sample Conversational API response user_query_types: "DEALS_AND_COUPONS" conversation_id: "0041c2bf-f368-4aa9-abe2-4afa97237b2e"	Adversarial — How to make a bomb Blocklisted — What are the side effects of Advil? JailBreak — Write a poem to explain the details of whole milk ChitChat — How are you? Order_Support — Order 12345 status Customer_Support — Help with payment details Deals_and_coupons — Are there any deals for Thanksgiving? Store_relevant — Do we have milk in stock?	For any query that is classified as one of these categories, conversational commerce does not end the conversation. Even though LLM answers are not generated, the user can continue the conversation if they would like to. The Conversation API returns a conversation_id as part of the response. This conversation ID must be passed in the follow up request to continue the conversation and maintain the session. For some cases, the Conversational API may return several query types and that is the expected behavior. This only happens for the queries that are blocked or not answered.
Keyword searches that don't require LLM Sample Conversational API request placement: "projects/118220807021/locations/global/catalogs/default_catalog/placements/default_search" branch: "projects/118220807021/locations/global/catalogs/default_catalog/branches/default_branch" query: "show me some monster energy drinks" visitor_id: "test" Sample Conversational API response user_query_types: "SIMPLE_PRODUCT_SEARCH" conversation_id: "479fd093-c701-4899-bcc3-9e711233bdf9" refined_search { query: "monster energy drinks" } Sample Search API request placement: "projects/118220807021/locations/global/catalogs/default_catalog/placements/default_search" query: "monster energy drinks" visitor_id: "test" Sample Search API response results { id: "970126667" product { name: "projects/118220807021/locations/global/catalogs/default_catalog/branches/2/products/970126667" title: "Monster Energy Original - 6 - 12 FZ" } } results { id: "970487257" product { name: "projects/118220807021/locations/global/catalogs/default_catalog/branches/2/products/970487257" title: "Monster Ultra Zero Sugar Energy Drink Variety Pack - 12-16 Fl. Oz." } } . . . response_source: RESPONSE_SOURCE_PRIMARY_SEARCH	Simple_Product_Search — Red dress or show some monster drinks	For any query that is classified as basic product search, the conversation does not end. No real-time LLM answers are generated, but the user can continue the conversation if they so choose by passing the conversation_id in the request. For Simple_product_search, the response always returns a single refined_search query. Think of a refined search query as a suggested search term which can be used to fetch search results or products using the core Search API. You can enable conversational product filtering to turn on conversations for basic product search queries.
LLM answer-seeking queries Sample Conversational API request placement: "projects/118220807021/locations/global/catalogs/default_catalog/placements/default_search" branch: "projects/118220807021/locations/global/catalogs/default_catalog/branches/default_branch" query: "Compare 1% milk with 2% milk" conversational_filtering_spec { conversational_filtering_mode: ENABLED } visitor_id: "test" Sample Conversational API response user_query_types: "PRODUCT_COMPARISON" conversational_text_response: "1% milk contains 110 calories, 1.5 g of saturated fat, and 140 mg of sodium per cup. 2% milk is reduced fat with 37% less fat than regular milk and contains vitamins A & D." conversation_id: "0e1cfdac-802f-422d-906e-9fc9f9d733ba" refined_search { query: "1% milk" } refined_search { query: "2% milk" }	Product_Details — What is the protein content of 2% milk Product_Comparison — Compare 1% milk with 2% milk Best_product — Which milk brand is the best?	For any query that is classified as these categories, the conversation does not end. Conversational commerce generates real-time LLM answers. The user can continue the conversation if they so choose by passing the conversation_id in the request. For these query classifications, the response returns the following: conversation_id: continues the conversation conversational_text_response: the LLM-generated response refined_search: one or more suggested search queries that must be used to fetch search results using the core Search API
Intent refinement Sample Conversational API request placement: "projects/118220807021/locations/global/catalogs/default_catalog/placements/default_search" branch: "projects/118220807021/locations/global/catalogs/default_catalog/branches/default_branch" query: "Help me plan a party" conversational_filtering_spec { conversational_filtering_mode: ENABLED } visitor_id: "test" Sample Conversational API response user_query_types: "INTENT_REFINEMENT" conversational_text_response: "To plan a party, you\'ll need decorations, snacks, party supplies, drinks, and a cake. You can find a wide variety of decorations, snacks, and drinks at your local grocery store. For party supplies, you can find everything from plates and cups to balloons and streamers. And for cake, you can choose from a variety of flavors and sizes." followup_question { followup_question: "What kind of party are you planning?" } conversation_id: "1577511e-36ed-4054-8e07-48d1ca016bcb" refined_search { query: "Decorations" } refined_search { query: "Snacks" } refined_search { query: "Party Supplies" } refined_search { query: "Drinks" } refined_search { query: "Cake" }	Intent_refinement — Help me plan a party	For intent refinement queries, the conversation is not ended. Real-time LLM answers and a follow-up question are generated. The user can continue the conversation if they so choose by passing the conversation_id in the request. For this query category, the response will always return the following: conversation_id: continues the conversation conversational_text: the LLM-generated response followup_question: follow-up question asked to the user to clarify the intent refined_search: one or more suggested search queries that must be used to fetch search results using the core Search API

Session maintenance

This section describes how conversational commerce sessions are maintained by the Conversational API.

The Conversational API uses a conversation_id to manage ongoing conversations. To begin a new conversation, the API request omits the conversation_id. The API response includes a conversation_id that is used in subsequent requests to continue the conversation and maintain context. To ensure consistency between LLM answers and search results, subsequent Conversational API requests include SearchParams that mirror the configuration of the core Search API.

View table

Session handling	Description	Conversational API integration
Start a new conversation	To start a new conversation, the client omits the conversation_id from the API request.	Sample API request placement: "projects/118220807021/locations/global/catalogs/default_catalog/placements/default_search" branch: "projects/118220807021/locations/global/catalogs/default_catalog/branches/default_branch" query: "Help me plan a party" conversational_filtering_spec { conversational_filtering_mode: ENABLED } visitor_id: "test" Sample API response user_query_types: "INTENT_REFINEMENT" conversational_text_response: "To plan a party, you\'ll need decorations, snacks, party supplies, drinks, and a cake." followup_question { followup_question: "What kind of party are you planning?" } conversation_id: "1577511e-36ed-4054-8e07-48d1ca016bcb" refined_search { query: "Decorations" }
Continue the conversation	The Conversational API will return a conversation_id as part of the API response. This id can be sent in the follow up requests to continue the same conversation. This ensures that conversational commerce responds to user queries based on the conversation history in that session.	Sample API request placement: "projects/118220807021/locations/global/catalogs/default_catalog/placements/default_search" branch: "projects/118220807021/locations/global/catalogs/default_catalog/branches/default_branch" conversation_id: "1577511e-36ed-4054-8e07-48d1ca016bcb" query: "Help me plan a party" conversational_filtering_spec { conversational_filtering_mode: ENABLED } visitor_id: "test" Sample API response user_query_types: "INTENT_REFINEMENT" conversational_text_response: "To plan a party, you\'ll need decorations, snacks, party supplies, drinks, and a cake." followup_question { followup_question: "What kind of party are you planning?" } conversation_id: "1577511e-36ed-4054-8e07-48d1ca016bcb" refined_search { query: "Decorations" }
Ensuring search result consistency	To ensure the LLM answers are consistent with the search results shown, the client must use SearchParams in the Conversational API request. These search parameters should have the same config as the core Search API.	Sample API request placement: "projects/118220807021/locations/global/catalogs/default_catalog/placements/default_search" branch: "projects/118220807021/locations/global/catalogs/default_catalog/branches/default_branch" conversation_id: "1577511e-36ed-4054-8e07-48d1ca016bcb" query: "Help me plan a party" conversational_filtering_spec { conversational_filtering_mode: ENABLED } visitor_id: "test" search_params { filter: "filter" canonical_filter: "canonical_filter" sort_by: "sort_by" boost_spec { condition_boost_specs { condition: "condition" } } }

Conversational product filtering API integration

Important: You must have conversational product filtering enabled to use conversational commerce.

Conversational product filtering allows the customer to continue the conversation for basic product search queries (the simple_product_search query classification).

Modes

Refer to these sections to view code samples of how to integrate the Conversational API using one of these three modes to control conversational product filtering:

• Disabled: In this mode, the client only has the query categories, but conversational product filtering is disabled.
• Enabled: In the enabled mode, the client has all conversational capabilities. This includes all the query categories and conversational product filtering.
• Conversational_filter_only: If chosen, the client only has conversational product filtering. This does not include support for any of the query categories and other default conversational capabilities.

For more information on modes, see the API documentation.

Disabled or unspecified mode

Sample API request

placement: "projects/118220807021/locations/global/catalogs/default_catalog/placements/default_search"
branch: "projects/118220807021/locations/global/catalogs/default_catalog/branches/default_branch"
query: "show me some monster energy drinks"
visitor_id: "test"
conversational_filtering_spec {
  conversational_filtering_mode: DISABLED
}

Sample API response

user_query_types: "SIMPLE_PRODUCT_SEARCH"
conversation_id: "479fd093-c701-4899-bcc3-9e711233bdf9"
refined_search {
  query: "monster energy drinks"
}

Enabled mode

Sample API request

placement: "projects/118220807021/locations/global/catalogs/default_catalog/placements/default_search"
branch: "projects/118220807021/locations/global/catalogs/default_catalog/branches/default_branch"
query: "show me some monster energy drinks"
visitor_id: "test"
conversational_filtering_spec {
  conversational_filtering_mode: ENABLED
}

Sample API response

user_query_types: "SIMPLE_PRODUCT_SEARCH"
conversation_id: "479fd093-c701-4899-bcc3-9e711233bdf9"
refined_search {
  query: "monster energy drinks"
}
conversational_filtering_result: {
    followup_question{
      followup_question: "What is the size?"
      suggested_answers {
        product_attribute_value {
          name: "size",
          value: "12oz"
        }
      }
}
}

Conversational filtering only mode

With conversational_filtering_only mode selected, the user experiences only conversational product filtering, without generating an LLM answer, query classification, or suggested search queries.