Semantic vector search in AI Search

  • Release version: Washingtondc
  • Updated July 16, 2025
  • 3 minutes to read
    • Summarize
    Summarized using AI
    This content was generated using new OpenAI-powered functionality. Results are provided on an as is basis and are not guaranteed to be accurate or complete.

    Summary of Semantic Vector Search in AI Search

    Semantic vector search enhances the Now LLM Service by finding results based on the meaning of search terms rather than just keyword matches. This feature is integrated into Now Assist Q&A Genius Results and Now Assist in Virtual Agent, improving how search queries are interpreted and matched to relevant records.

    Show full answer Show less

    Key Features

    • Semantic vector search analyzes the meanings and context of search terms, providing results that reflect the user's intent.
    • Search results are ordered by similarity to the search query, rather than strictly on keyword matches.
    • It operates automatically without the need for configuration or maintenance of synonym dictionaries.

    Key Outcomes

    • Improves recall for Knowledge articles, Catalog Items, and topics by matching results based on meaning.
    • Delivers a more natural search experience, returning relevant results for varied queries with similar meanings.
    • Reduces reliance on keyword-based search, providing more accurate and contextually relevant results.

    Availability

    Semantic vector search is available in several contexts including:

    • Now Assist in Virtual Agent for Catalog Item and live agent topic retrieval.
    • Now Assist Q&A Genius Results for Knowledge articles and cached answers.
    • Now Assist Actions Genius Results for Catalog Items.

    Note that semantic vector search currently supports English only.

    Semantic vector search allows the Now LLM Service to find results based on how similar they are in meaning to your search terms. Now Assist Q&A Genius Results and Now Assist in Virtual Agent use semantic vector search to improve recall with natural language interpretation of search queries.

    By default, AI Search uses lexical search, meaning that it finds results for records that contain literal keyword matches for the terms in your search query. Starting in the Vancouver Patch 4 release, AI Search includes an alternate search mode called semantic vector search that's used in Now Assist Q&A and Now Assist Actions Genius Results.

    Semantic vector search analyzes the meanings and context of your search terms and uses that information to find results with similar meanings. It improves search recall by interpreting natural language to more accurately reflect your search's intent.

    As an example, suppose you index a source record including text how to prevent phishing and search for avoiding scams.
    • In lexical search mode, AI Search returns no result for this record because your search query terms aren't literal matches for the terms included in the indexed content.
    • In semantic vector search mode, however, AI Search might return a result for the record based on the following context-aware analysis:
      • Your search term avoiding is contextually similar in meaning to prevent
      • Your search term scams has meaning that overlaps with the meaning of phishing

    AI Search orders results from semantic vector search based on how similar they are to your search. In the example, if you had a second indexed record including text how to prevent scams, it would be more similar to the avoiding scams search and its search result would appear before the how to prevent phishing result.

    Unlike the default keyword-based search mode, semantic vector search does not rely on your synonym dictionaries to find term equivalences. AI Search identifies term similarities automatically when indexing source content and metadata for semantic vector search.

    Semantic vector search is automatically enabled and doesn’t include any configurable settings.

    Benefits of semantic vector search

    In comparison with the default keyword-based lexical search mode, semantic vector search provides the following benefits:
    • Helps improve recall for Knowledge article, Catalog Item, and topic retrieval by matching results based on the underlying meaning of your search rather than matching on keywords.
    • Provides a more natural and consistent search experience by returning similar results for varied search queries with the same underlying meaning.
    • Reduces the need to create and maintain synonym dictionaries to anticipate similarities in searches.

    Semantic vector search output

    Semantic vector search overrides the normal AI Search term matching and relevancy ranking mechanisms. Results are ordered based on their computed similarity to your search query terms, with the most similar results appearing first.

    Availability of semantic vector search

    Semantic vector search is available in the following contexts.
    • Now Assist in Virtual Agent uses semantic vector search for Catalog Item retrieval and live agent topic retrieval. For more details, see Now Assist in Virtual Agent.
    • Now Assist Q&A Genius Results use semantic vector search along with legacy keyword search when looking for Knowledge articles that match your search query. They also use semantic vector search when looking for cached answers that match your query in the second-level cache. For more details, see Now Assist Q&A Genius Results and Caching for Now Assist Q&A Genius Results.
    • Now Assist Actions Genius Results use semantic vector search along with legacy keyword search when looking for Catalog Items that match your search query. For more details, see Now Assist Actions Genius Results.

    Semantic vector search only supports searches in English.