Hybrid search in AI Search
Summarize
Summary of Hybrid search in AI Search
Hybrid search in AI Search combines traditional keyword search with semantic vector search to deliver results that best match both the terms and the underlying meaning of user queries. While keyword search matches exact terms, semantic vector search interprets the context and intent behind those terms, improving recall and relevance.
Show less
Starting with the Zurich Patch 9 release and the installation of Now Assist in AI Search, hybrid search uses Reciprocal Rank Fusion (RRF) to blend keyword relevance scores and semantic similarity scores into a unified ranking. This hybrid approach balances precision from keyword matching with contextual understanding from semantic search.
By default, hybrid search applies to indexed data sources that are included in the semantic index, such as Knowledge Table, Catalog Item Table, Skills, and External Content Connectors. Additional sources can be manually configured for inclusion.
Benefits
- Improved search quality and recall: By combining keyword and semantic matching, hybrid search retrieves more relevant results, even if queries are phrased differently.
- More relevant top-ranked results: Semantic understanding helps surface answers aligned with the user’s intent, not just keyword matches.
- Fewer zero-result searches: The semantic component can interpret vague or misspelled queries, reducing no-result outcomes.
Availability and Activation
Hybrid search is available from the Zurich Patch 9 release onward and requires the Now Assist in AI Search application. For new installations of Now Assist in AI Search 17, hybrid search is enabled automatically for configurations using AI Search as their engine.
Legacy search configurations using the Zing text indexing engine do not support hybrid search. For upgrades, hybrid search must be manually activated via the AI Search Admin console.
Interactions and Considerations
- Activating hybrid search disables search result counts for facets within the affected search application configuration.
- The “Most recent” sort option behavior changes: AI Search merges keyword and semantic results before sorting by last modification date.
Next Steps
To leverage hybrid search’s improved precision and recall in your ServiceNow environment, ensure Now Assist in AI Search is installed and configure your search application settings to enable hybrid search, especially for knowledge articles, catalog items, external content, and topic retrieval.
In hybrid search mode, AI Search blends keyword search and semantic vector search to find results that best match the terms and meaning of your search.
Overview of hybrid search
By default, AI Search performs most searches in keyword search mode. This mode looks for the best matches for your search terms, but doesn't take the context or meaning of those terms into account. The keyword search relevance score for a search result indicates how well the indexed terms in that search result match your search terms.
Starting in the Yokohama Patch 11 release, AI Search includes an alternate semantic vector search mode in features which work with the Now LLM Service. 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. The semantic vector search relevance score for a search result indicates how closely the search result matches the meaning of your search.
- Knowledge Table indexed source [kb_knowledge table]
- Catalog Item Table indexed source [sc_cat_item table]
- Skills indexed source [sys_gen_ai_skill table]
- All External Content Connectors indexed sources [connector-specific tables]
You can manually configure additional indexed sources to be included in the semantic index, extending hybrid search to those sources as well.
To learn more about semantic vector search mode and how it differs from keyword search mode, see Semantic vector search in AI Search.
Benefits of hybrid search
- Improved search quality and recall
- Hybrid search combines keyword and semantic matching, increasing the chances of retrieving relevant results, even when users phrase their queries differently.
- More relevant top-ranked search results
- Semantic understanding helps surface answers that better match the user’s intent instead of just their search keywords.
- Fewer zero-result searches
- Hybrid search can return useful results even for vague or misspelled queries because it uses semantic matching to understand meaning beyond exact keywords. This reduces the number of searches that return no results.
Availability of hybrid search
Hybrid search is available beginning with the Zurich Patch 9 release. It requires installation of the Now Assist in AI Search application.
If you upgrade to Now Assist in AI Search from a previous version, hybrid search isn't automatically activated for your search application configurations that use AI Search as their search engine. To learn how to manually activate hybrid search for these search application configurations in the AI Search Admin console, see Manage hybrid search in search applications.
Interactions with other features
Activating hybrid search in a search application configuration disables search result counts for facets in that search application configuration. For more information on search result counts for facets, see Show search result counts for facets on the results page for a search application.
When you activate hybrid search in a search application configuration, the behavior of the Most recent search result sort option changes for the specified search application. Instead of sorting the entire list of search results by last modification date, AI Search first retrieves keyword search results and the most relevant semantic vector search results, then merges those results and sorts them by last modification date. To learn about sorting search results based on recency, see Change the sort order for your search results.