---
title: Conversational Product Discovery
slug: q2pb2427
canonical_url: https://docs.coveo.com/en/q2pb2427/
collection: coveo-for-commerce
source_format: adoc
---
# Conversational Product Discovery

Coveo Conversational Product Discovery is a search-native conversational experience that dynamically blends products, content, and concise explanations.
Think of it as a tool that lets shoppers describe what they need in their own words and receive curated product results directly within your existing search experience.

From a single search entry point, shoppers can:

* Express shopping needs in natural language

* Receive curated product results

* Refine product results through follow-up queries

* Compare products and characteristics

* Build relevant product bundles

* Learn more about returned products

Unlike standalone chatbots that fragment user journeys, Conversational Product Discovery unifies conversation and search to preserve speed, relevance, and monetization.
The experience is powered by AI agents grounded in your [catalog data](https://docs.coveo.com/en/obcf0333/), constrained by enterprise-defined layouts and guardrails, and rendered through an adaptive UI that delivers responsive experiences across devices.

> **Important**
>
> Conversational Product Discovery must be enabled for your [Coveo organization](https://docs.coveo.com/en/185/).
> Contact your Coveo representative to get started.

## Prerequisites

Before enabling Conversational Product Discovery, ensure you meet the following prerequisites:

* Have an active Coveo for Commerce license that includes the Conversational Product Discovery feature.

* Conversational Product Discovery has been enabled for your [Coveo organization](https://docs.coveo.com/en/185/) by your Coveo representative.

* Have [indexed your catalog data](https://docs.coveo.com/en/3448/)

* Have configured a [catalog entity](https://docs.coveo.com/en/3139/) and make sure that:

** The [standard commerce fields](https://docs.coveo.com/en/n73f0502/) are populated and mapped in your catalog entity.

** Configured a [storefront association](https://docs.coveo.com/en/o48e0216/) to make the products available in the [catalog entity](https://docs.coveo.com/en/3143/) accessible via the Commerce API.
A storefront association automatically generates a [query pipeline](https://docs.coveo.com/en/180/) for handling queries from the associated storefront, which is required for Conversational Product Discovery.

* [Track commerce events](https://docs.coveo.com/en/3188/) on your storefront.

## What you can do with Conversational Product Discovery

Conversational Product Discovery addresses a growing gap in digital commerce: shoppers increasingly express complex, exploratory intents that traditional keyword search alone can't satisfy.
On that foundation, Conversational Product Discovery:

* **Converts exploratory traffic into revenue** by guiding shoppers from vague needs to curated, purchase-ready product selections.

* **Preserves transactional search performance** by augmenting, not replacing, your existing high-performing search experience.

* **Reduces dead-end experiences** by always offering follow-up suggestions, refinements, and next-best actions.

* **Maintains merchandising control** through deterministic layouts, guardrails, content exclusions, and orchestration directives that constrain AI agent behavior to business-defined outcomes.

Conversational Product Discovery supports six types of conversational interactions:

[cols="1,2,2", options="header"]
|===
|Interaction type
|Description
|Example

|Single-intent product search
|Shoppers describe what they want and receive a curated list of relevant products.
|"I want to see yellow paddleboard options under $1000."

|Multi-intent product search
|Shoppers request products across multiple categories in a single query, receiving distinct product lists per category.
|"Show me wetsuits, a bag, and surfboards."

|Conversational refinement
|Shoppers refine returned results through follow-up messages while Conversational Product Discovery maintains full context.
|"Under $1000."

|Product education
|Shoppers learn about products through catalog content and product attributes.
|"Is this bag waterproof?"

|Product comparison
|Shoppers compare products using a structured table that highlights relevant attributes.
|"What are the differences between these 4 surfboards?"

|Product bundle
|Shoppers receive a curated bundle of complementary products across categories to meet a stated goal.
|"What do I need to get started surfing? I have a $400 budget."
|===

## Core building blocks

Conversational Product Discovery is built on three core building blocks:

* [**Layouts**](https://docs.coveo.com/en/q2pb2437/) define the types of conversational interfaces Conversational Product Discovery can render.
Each layout corresponds to a detected intent type and specifies the UI composition, constraints, and rendering rules that govern the response.
Layouts give merchandisers deterministic control over the conversational experience while still allowing flexibility to adapt to shopper needs.

* The [**discovery agent**](https://docs.coveo.com/en/q2pb2442/) is the central orchestrator.
It interprets shopper input, detects intent, selects the appropriate layout, retrieves products from the Commerce API, assembles a grounded response, and streams the UI structure to the front end.
The agent operates through a policy-driven execution model with configurable guardrails, content exclusions, and orchestration directives.

* The [**adaptive canvas**](https://docs.coveo.com/en/q2pb2447/) is the agent-controlled UI surface that dynamically renders the experience.
Using the A2UI (Agent-to-UI) framework, the agent returns declarative component descriptions (carousels, comparison tables, action buttons) that the front end interprets and renders as native widgets.
The canvas adapts to desktop, mobile, and tablet.

![The core building blocks of Conversational Product Discovery | Coveo](https://docs.coveo.com/en/assets/images/coveo-for-commerce/images/cpd-building-blocks.png)

## End-to-end flow

When a shopper interacts with Conversational Product Discovery, the following sequence occurs:

. **The shopper enters a query**:
The shopper types a natural language message into the search experience.
The query can be a product search, a multi-category request, a follow-up refinement, a question about a product, a comparison request, or a bundle inquiry.

. **The discovery agent interprets the input**:
The agent analyzes the message to identify the underlying intent.
It leverages conversational context from previous interactions within the session to understand follow-up questions, refinements, and references to earlier results.

. **The agent selects a layout**:
Based on the detected intent and any configured rules, the agent selects the most appropriate [layout](https://docs.coveo.com/en/q2pb2437/).
When a query could match multiple layout types, the agent resolves to the most specific match.
For example, a query that contains both a product search and a comparison request is handled as a comparison rather than a simple product search.

> **Note**
>
> Merchandising rules, such as layout override, take precedence over AI-driven classification.

. **The agent calls tools to retrieve data**:
The agent calls the appropriate tools to retrieve the information required to assemble a response.
It currently uses the Commerce API as its primary tool for retrieving product information from your indexed catalog data.


The agent may execute multiple retrieval passes to optimize results, for example broadening a query that returned too few results, or narrowing one that returned too many irrelevant products.

. **The agent assembles the response**:
The agent constructs a response that satisfies the detected intent while adhering to the selected layout's constraints on content order, product grouping, copy, and presentation format.

. **The agent streams the UI**:
The agent generates a declarative UI structure and streams it progressively to the front end, where the [adaptive canvas](https://docs.coveo.com/en/q2pb2447/) renders it as native UI components.

## Policy-driven execution

The discovery agent enforces deterministic behavior through a structured execution model.
Each step in the flow operates with its own context, model selection, and access controls:

* **Optimized model selection**:
Different steps can use different AI models.
Steps that require fast routing use lightweight models, while steps that require nuanced synthesis use more capable models.

* **Controlled access**:
Only the tools and data sources authorized for the current step are available to the AI model, preventing unintended actions.

* **Execution enforcement**:
A policy engine validates every action and step transition, preventing the AI model from deviating from the defined execution flow.

## Multi-turn context

The discovery agent maintains conversational context across interactions.
Shoppers can refine results, ask follow-up questions, and reference earlier products without restating their original query.

* For anonymous shoppers, context is maintained at the session level.

* For logged-in shoppers, conversation history can persist beyond sessions with a configurable time-based expiration.

**Example**

A shopper searches for "paddleboards" and receives a curated product list.
They then follow up with "under $1000".
The agent refines the same result set without the shopper needing to repeat "paddleboards."

Next, they ask "is this one good for beginners?" referring to a specific product.
The agent uses conversational context to identify the referenced product and provides education content drawn from catalog data.

## Grounding and accuracy

The discovery agent is grounded exclusively in your product catalog data accessed via the Commerce API.
The agent doesn't invent products or fabricate product attributes.
All product information in responses comes directly from your indexed catalog data.
Content guardrails prevent the agent from revealing system instructions, inventing product data, or straying into blocked topics, words, or brands that you define.

![Diagram imagining grounding and accuracy](https://docs.coveo.com/en/assets/images/coveo-for-commerce/images/grounding-and-accuracy.png)

> **Note**
>
> Conversational Product Discovery uses generative AI models.
> While responses are grounded in your product catalog and constrained by enterprise-defined layouts and guardrails, AI-generated content such as rationale text and educational bullets may occasionally contain inaccuracies.
> Review the experience configuration and monitor responses, particularly during initial deployment.

## What's next

* [Layouts](https://docs.coveo.com/en/q2pb2437/): Learn how layouts define the conversational experience.

* [Discovery agent](https://docs.coveo.com/en/q2pb2442/): Understand how the agent orchestrates discovery journeys.

* [Adaptive canvas](https://docs.coveo.com/en/q2pb2447/): Learn how the agent-controlled UI renders the experience.