This feature is in beta, and we are actively working to improve it. If you have any questions, please contact our Support.

This guide describes how to integrate product recommendations to headless stores using the VTEX Recommendations BFF API.

Before implementing headless recommendations, make sure you have:

• A frontend capable of adding third-party scripts and making authenticated HTTPS requests
• Requested and received approval for the product recommendations feature setup from VTEX Support.

After you request and receive approval for using the product recommendations feature, the VTEX Support team will prepare your environment by:

• Creating a workspace for recommendations
• Synchronizing your product catalog with the recommendation service
• Setting up API keys for secure access
• Generating a custom integration script for your store.

Once these configurations are complete, you can proceed with the implementation steps below.

After receiving the integration script from VTEX Support, implement it in your headless store:

1. Create a dedicated component or module in your codebase for third-party integrations.
2. Implement the script based on your framework's best practices for third-party script management. Check an example of implementation in FastStore: Third-party Scripts
3. Ensure the script runs on all pages where you want to enable recommendations or track user behavior.

The script handles:

• User tracking initialization
• Browsing and behavioral data collection
• Model training data gathering.

4. Save the _snrs_uuid in the user's order form, so that when the order is placed, you can associate that sale with the user identified during navigation.

When the ID is generated, you must save the user's ID in the orderForm during navigation using PUT Set multiple custom field values as shown below:

To validate that it worked, request to retrieve the orderForm using GET Get current cart. The customData field returned should contain the added information.

5. Send the POST Product View event following the API reference.

Once the tracking script collects enough browsing data, the VTEX Support team will train and configure the recommendation models. The available strategies are:

Name	Description
Most popular	Products with the highest number of views
Recommended for you	Custom recommendations based on user profile and behavior
Similar products	Products similar to a specific one
Cross-sell	Complementary products for a specific one that are commonly purchased together.
Best sellers	The store's best-selling products
Last seen	Products recently viewed by the user
Recent interactions	Products most likely to engage the user in the future
Visually similar products	Products visually similar to a specific one
Manual collection	Recommendations from a collection created manually

After the models are trained, they will be available for API-based recommendation displays in your store.

To implement product recommendations in your headless store, follow these steps:

1. Start a user session to get a unique identifier
2. Fetch recommendations for your shelves.
3. Track user interactions with the recommendations.

Use the same user identifier (recommendationsUserId) across all requests in a session.

Use the POST Start Session endpoint to create a unique user ID for recommendations. This endpoint:

• Creates a unique user ID (recommendationsUserId) for the session.
• Links it to the provided order form ID.
• Stores the ID in the vtex-rec-user-id cookie for session continuity.
• Initializes recommendation tracking and personalization.

Call this endpoint on the first page view of each visit, before displaying any recommendations, and only if the _snrs_uuid is not already present. If orderFormId is not provided in the request body, the API attempts to retrieve it from the checkout.vtex.com cookie.

Request example:

Response example:

Persist the returned recommendationsUserId as _snrs_uuid and include it in all subsequent requests.

Use the GET Fetch recommendations endpoint to fetch products for each shelf whenever a page contains a recommendation component (homepage, PDP, cart, PLP, etc.).

The campaign determines the type of recommendations returned, such as personalized recommendations, similar items, cross-sell, and others.

This endpoint retrieves a list of recommended products based on:

• Campaign VRN (Virtual Resource Name) identifying the recommendation strategy. The VRN follows the pattern vrn:recommendations:{store-name}:{campaignType}:{campaignId}. Contact our Support to obtain the campaignId

  Available campaign types:

  • rec-top-items-v2: Best sellers
  • rec-persona-v2: Personalized recommendations
  • rec-similar-v2: Similar items
  • rec-cross-v2: Cross-sell
  • rec-cart-v2: Cart-based recommendations
  • rec-last-v2: Last seen
  • rec-interactions-v2: Recent interactions
  • rec-visual-v2: Visual similarity
  • rec-search-v2: Search-based recommendations
  • rec-next-v2: Next interaction

• User context (from recommendationsUserId)

• Page context (product ID, cart items, etc.).

Request example:

Response example:

Render the returned items as product cards on the shelf. After rendering, proceed to event tracking.

To improve recommendation accuracy and measure effectiveness, you need to track how users interact with the recommended products. There are three types of events to track:

• Recommendation views: When recommendation shelves become visible to users.
• Recommendation clicks: When users click recommended products.
• Product views: When users visit product detail pages.

⚠️ For accurate tracking, ensure that:

• All events include the same recommendationsUserId.
• Events are properly debounced to prevent duplicates.
• View events are triggered only when shelves enter the viewport.

Recommendation views

Use the POST Recommendation view endpoint when a recommendation shelf becomes visible. This endpoint:

• Records when recommendations are shown to users.
• Provides data for click-through rate (CTR) analysis.
• Helps train and improve recommendation models.
• Should be called when shelves enter the viewport.

Request example:

Recommendation clicks

Use the POST Recommendation click endpoint when users interact with recommended products. This endpoint tracks user engagement with recommendations and should be called for each click on a product card.

Request example:

Product views

Use the POST Product View endpoint when users view product details. This endpoint tracks all product detail page views.

⚠️ VTEX automatically tracks product views once the Recommendations feature is set up in your store’s website as instructed in Step 2. This event is only required for app implementations or other channels. Track product views for all sources to ensure comprehensive data collection for the recommendation models.

Request example:

• Recommendations BFF API Reference
• Product Recommendations Overview