Note: The feature previously known as AI Search is now referred to as AI Semantic Search. This update reflects a shift to a large language model (LLM)-based system.
The information in this document remains accurate for both AI Semantic Search and the earlier AI Search RNN (legacy) version. If your implementation still uses the RNN-based model, you can continue to rely on the details provided here. Functionally, the concepts and integrations described have not changed.

In the Merchandising Studio, search behavior is defined by a search configuration (also referred to as "search profile"). This configuration is made up of one or more search passes, which are ordered stages that determine how search results are retrieved.

Each pass in the configuration searches for product matches using different criteria, ranging from exact product ID matches to more descriptive data, e.g. product names or attributes. The system processes each search pass in order. Whether or not all search passes are activated depends on the overall configuration, i.e. it is possible that search passes ranked lower are skipped if the earlier passes have already led to a sufficient number of results.

The role of AI Semantic Search in the configuration

AI Semantic Search is integrated as an additional, supportive search pass. 

AI Semantic Search is usually placed near the bottom of the configuration. Its purpose is to complement Fredhopper Standard Search by acting as a fail-safe when Standard Search doesn’t return results. Standard Search is highly configurable and typically very effective. It searches structured product data such as, product names, categories, attributes and more.

Search passes are carefully ordered and weighted to ensure high relevance. For instance, if a customer types "red short sleeve dress," Standard Search can accurately match those words to product attributes and return highly relevant results.

If Standard Search fails to return results, especially for long-tail queries or more natural, expressive searches like "nice shirt for a summer wedding", AI Semantic Search steps in. It analyzes historical search behavior, i.e. search terms linked with clicked products (for more information see Preparing for AI Semantic Search), to find relevant matches, even when the search term doesn’t exist in the product data.

Warning: While you can view and access your search configuration in the Merchandising Studio, we strongly advise against making changes without consulting your Crownpeak Technical Consultant. Unsupervised adjustments, such as altering search passes or field weightings, can unintentionally impact search performance and relevance.

If changes are needed, always test carefully in a non-production environment and consult your Crownpeak Technical Consultant to ensure the configuration continues to deliver high-quality results.

Displaying AI Semantic Search Results

Once results are retrieved from a search, Ranking Cocktails are used to order them. Ranking Cocktails can be applied to both results retrieved from Standard Search as well as AI Semantic Search.

You can also apply result modifications to influence which AI Semantic Search results are displayed, e.g. hiding embargoed products or filtering by region. However, we generally do not recommend restricting AI Semantic Search results unless there's a specific business need.