Setting up an effective search configuration is essential to unlocking the full potential of Experience Search. The right settings ensure that you deliver a seamless and relevant experience for your users. This guide explains how to configure searchable fields, define user filters (also known as facets), and adjust other key settings that influence search behavior.
Searchable fields
Searchable fields tell Experience Search which product attributes to use when evaluating user queries — whether text or image-based. While the engine uses your entire product catalog, searchable fields guide it to prioritize the most relevant data.
By selecting the right searchable fields, you ensure the search experience highlights the product attributes that matter most to your business, leading to more accurate and meaningful results that improve user engagement.
In the Configuration screen, you can set up the following:
- Visual attribute: The image to evaluate for a search query.
- Textual attributes: Product feed column names to evaluate for a search query, including weighing their importance.
- Facets: Attributes users can filter by, including which values are displayed and how.
Visual attribute
Experience Search is a multimodal search engine, meaning it uses both visual and textual inputs to return relevant results. Visual data is central to its Visual Search and Semantic Search capabilities, and the visual attribute is at the core of this function.
By default, the image_url column is automatically selected. When you create a search campaign, Experience Search begins training on your product images using this default source. If your feed includes a more suitable image column, you can replace the default with a different image source at any time, to ensure that the search engine learns from the most relevant visuals for your products.
Note that you can select only 1 column.
Image quality guidelines
Image quality is critical for search performance — especially Visual Search. High-quality images help the visual model better understand and represent your products, leading to more accurate results.
To optimize performance, follow these best practices:
- Use studio-quality images of products against a plain white or grey background.
- Avoid additional objects or visual distractions in the frame.
These guidelines help ensure the model focuses entirely on the product itself, significantly improving the accuracy of visual search results.
Textual attributes
Experience Search analyzes both your product images and textual metadata to create a rich, structured representation of each item. These representations — derived from both visual and textual signals — enable highly accurate, relevant search results. By selecting the right textual attributes, you guide Experience Search to prioritize the metadata that matters most to your brand and customers.
Recommended searchable fields
For the best performance, we recommend using fields that contain commonly-used keywords and terms:
- Product name/title
- Categories or product type
- Brand
- Key attributes (such as color, material)
Best practices, technical requirements, and limitations
Follow these guidelines to ensure strong search performance:
- Only the STRING data type is supported (categories and keywords are ARRAYs, but allowed).
- Don't use long-form content like product descriptions, SEO text, or marketing copy.
Dynamic Yield creates a a description-like internal field as part of the training process to handle natural language queries. - The maximum number of searchable fields is 15. However, we highly recommend using no more than 5. Using too many fields can introduce noise and reduce relevance.
- Irrelevant fields (not directly describing the product) might harm search performance.
- Only 2 high-priority fields can be defined for weighted importance.
- Adding fields beyond core attributes like Product Name and Category typically doesn't typically improve results.
Carefully selecting and prioritizing searchable fields tailors the search behavior to reflect your brand’s unique catalog and customer needs.
Localization for searchable fields
Experience Search is a multi-language search engine that enables you to deliver localized search results tailored to each market. There are two primary approaches to managing localization:
Option 1: Use separate sections per locale
Some retailers use dedicated sections for each locale. Benefits include:
- Smaller, more focused catalogs per region
- Independent performance reporting
- Regional-specific management and optimization
- Each section operates using a single language, optimizing relevance and control
Option 2: Localized columns in a single product feed
With this approach, you manage localization within a single product feed by including language-specific columns for each locale (for example, es_US:name, en_US:name, and so on). This method streamlines operations while supporting multiple languages from a single feed:
- A single-team managing all locales saves time and effort
- A unified feed structure provides consistency and simplifies updates
- Consolidate performance reporting into one centralized view for easier tracking and analysis
Important: In the single-feed localization approach, every searchable field must be localized, meaning it includes a language and country code (for example, en_US:name, es_US:category).
If a localized version of a searchable field is missing for a specific locale, Experience Search falls back to the default field.
If the localized field exists but is empty, it is ignored entirely in the retrieval process — and no fallback occurs.
Example:
- name is a searchable field.
- The product feed contains name and en_US:name.
- A search request comes from the es_US locale:
- If es_US:name doesn’t exist, the system falls back to name.
- If es_US:name exists but is empty, the field is ignored and no fallback is applied.
This ensures that only meaningful, populated data is used in the search process.
Facets
Facets are product attributes returned in the search response that support filtering, such as brand, color, size, or category. They enable site visitors to narrow down results based on relevant criteria.
In other words, facets translate product data into filters. This improves navigation, supports product discovery, and enhances the overall shopping experience.
Each facet includes the following components:
- Column name: The product feed column used as the source for the facet (brand, color, size).
- Display name: The label displayed on filter menu (“Brand”, “Color”, “Size”).
- Sort by: Facets can be ordered by count or in alphabetical order.
- Use for: Select to display in Visual Search, Semantic Search, or both.
Note: You can set up to 50 facets. Each facet can have up to 100 values for any given search query. Only facet values with a count greater than 0 are displayed.
Multi-select facets
To define a multi-select facet:
- Define an array type column in your product feed:
type:array:column_name - In each row, enter the values for each product separated by pipes (|)
For example, to create a color facet, add this column: type:array:color and enter values such as white|purple|black - Select the column as a facet in your Experience Search variation configuration.
Settings
Configure the following settings:
- Default language: Experience Search can understand queries in any language. To process and interpret the meaning of queries more effectively, you must configure a primary (default) language.
- Brand name: Enter your brand name to provide context that helps our algorithms improve search query translations.
Display only the most relevant product from each group: Experience Search operates in two modes.
By default, Experience Search displays only the most relevant product from each group. In this mode, it automatically selects a group leader — the product that best matches the user’s query.For example, if a user searches for “glamorous sparkling dress” and there is a collection of the same dress available in 10 colors (sharing the same group ID), Experience Search displays only the most relevant variant as the group leader.
If you turn off this mode, Experience Search might return multiple SKUs from the same group, as long as more than one product is relevant to the user’s query.