Vector Search

Overview

Vector search enhances the search experience by finding results based on meaning, not just exact keywords. It helps users discover relevant content even when traditional keyword search fails. Unlike keyword search, which relies on exact word matches, vector search represents text as high-dimensional vectors (also known as embeddings) that capture context and semantics. This allows the system to present results that align with the intent of the query, even when different words, synonyms, or natural phrasing are used.

How does it work?

• Text is converted into vectors: Both the user query and the searchable content are transformed into high-dimensional numeric representations called vectors. These vectors capture the meaning of the text.
• Similarity is measured: The system compares vectors and returns the results most similar to the query using a proprietary similarity metric.

Privacy & Technology Note

Powered by advanced vector-based language understanding but does not rely on generally available AI or LLMs (large language models). All processing happens on our servers; your data never leaves our infrastructure.

When Vector Search Helps

Vector search is especially valuable when:

• You want results that match intent, not just exact keywords.
• Queries are written in natural language, such as full sentences or questions.
• Users often search with synonyms, related terms, or varied terminology.
• Queries may contain typos, spelling errors, or other mistakes.
• Queries may include mixed languages.

Pros and Cons

Pros:

• Finds relevant results even with vague, unstructured, or error-prone queries.
• Handles synonyms and paraphrases automatically.
• Great for exploratory or conversational search experiences.

Cons:

• Results can be less explainable.
• Accuracy and relevance may require some fine-tuning.

Vector search is available in two modes:

• Fallback Vector Search – used when keyword search returns no results.
• Hybrid Search – combines keyword and vector search for broader coverage.

Prepare product data for Vector Search

Vector Search uses the meaning and context available in your product data. Result quality therefore depends on complete, accurate, and relevant product information.

Tuning cannot fully compensate for product data that is missing, misleading, or lacks context. Before evaluating or tuning Vector Search, review the data applicable to your catalog.

Pay particular attention to accurate product titles or names, concise relevant descriptions, product attributes, and search attributes or contextual and use-case tags. Use the fields that fit your catalog. Follow the documented required, recommended, and optional field guidance.

Pre-tuning checklist

• Check that titles accurately identify and distinguish products.
• Improve descriptive text, attributes, and search or context tags when they express product characteristics, intended use, or related terminology.
• Identify incomplete, misleading, or missing information that could reduce semantic context.
• Test representative natural-language customer queries before changing settings or judging relevance.

Enriching applicable data improves the semantic context available for relevance. Review Data Feed requirements and How to optimize your Datafeed field guidance.

Suggest

For all setups where Vector Search is enabled and active, vector search will also be automatically active in Suggest. This will ensure similar results in Suggest as in the product search.