---
sourceDocument: Zurich ServiceNow AI Platform Administration
sourceDocumentLink: https://www.servicenow.com/docs/r/zurich/platform-administration

 Release :

    - zurich

ft:locale :

    - en-US

ft:publication_title :

    - Zurich ServiceNow AI Platform Administration

ft:clusterId :

    - platadm

bundleId :

    - platadm

workflow :

    - Platform


---

# AI Search RAG (Retrieval-Augmented Generation)

# AI Search RAG (Retrieval-Augmented Generation) {#ariaid-title1}

Release version: Zurich  
Updated October 13, 2026  
![](https://www.servicenow.com/docs/portal-asset/ico-clock) 2 minutes to read
Summarize  
![AI sparkle icon](https://servicenow.com/docs/portal-asset/ai-sparkle-icon) 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 AI Search RAG (Retrieval-Augmented Generation)

The AI Search RAG (Retrieval-Augmented Generation) capability in ServiceNow enhances AI Search accuracy by focusing large language models (LLMs) on specific, contextual data sets rather than broad general data.
This technology combines information retrieval with text generation to deliver more relevant and context-rich search results.
Show full answer Show less  

## Key Features

* **Indexing:** Organizes and prepares data into searchable formats using the `aisdatasource` table and semantic indexing configurations, enabling efficient retrieval.
* **Search Process:** Utilizes the RAG retrieval API to find relevant results within indexed data based on user queries.
* **Core Components:**
  * **Query Rewrite:** Simplifies complex or lengthy queries to avoid token limits in embedding models.
  * **Retriever:** Extracts the most relevant results from indexed sources using the rewritten query.
  * **Re-ranker:** Applies LLM to reorder and refine search results for maximum relevance.
  * **Final Response:** Delivers the refined answer to the user as the search output.
* **Embedding Model:** Uses semantic and vector search techniques with embedding models to map queries and data for precise retrieval. The default model is ServiceNow Embedding (E5), with support for third-party embedding models (e.g., Azure OpenAI Embedding, Gemini Text Embedding) and custom hosted models.
* **Activation:** The AI Search RAG functionality is enabled through the `snaisrag` plugin, which activates automatically with the installation of Generative AI Controller or any ServiceNow Otto® application.

## Key Outcomes

By implementing AI Search RAG, ServiceNow customers can expect:

* Improved search accuracy and relevance by leveraging focused, contextual data retrieval rather than broad, general datasets.
* Enhanced user query handling through query rewriting, ensuring compatibility with embedding model limits.
* More precise, context-aware responses generated by combining indexed data retrieval with advanced LLM re-ranking.
* Flexibility to use default or third-party embedding models, or custom models, allowing tailored search experiences.
* Seamless integration with existing AI Search applications through automatic plugin activation.  
The AI Search RAG (Retrieval-Augmented Generation) capability provides a framework that enhances the search accuracy of AI Search results by narrowing the focus of your large language model (LLM) to a specific, contextual data set, instead of the vast, general data it was trained on.

## AI Search RAG overview {#ais-rag__section_mjd_tfh_ggc}

RAG (Retrieval Augmented Generation) technology helps enhancing search capabilities of your AI Search application by combining information retrieval with text generation. It includes the following steps:

Indexing
:   Prepare and organize data to make it searchable. It includes the ais_datasource table and semantic indexing configurations. For more information, see [Indexed sources in AI Search](https://www.servicenow.com/docs/YBjHfOUBiGWYXcO4ejBRKA "Indexed sources designate ServiceNow AI Platform tables and external document sets with alphanumeric text and string field content that you want to make searchable. AI Search ingests text and string fields from table records or external documents and stores their searchable alphanumeric content in its search index.").

Searching
:   Find the most relevant results from the indexed data using a query. It includes RAG retrieval API to run search on indexed data of the search profile, and RAG capability that manages the RAG process, including retriever and
re-ranker to deliver relevant results.  
Components of AI Search RAG  
The AI Search RAG capability orchestrates the complete RAG process using components, such as query rewrite, retriever, and re-ranker to deliver the most relevant results.

* Query Rewrite: Rewrites the complex or long queries into concise, short form to avoid truncation issues with the embedding model's token limit.
* Retriever: Uses the rewritten query to retrieve the most relevant results from indexed sources.
* Re-ranker: Uses LLM to further refine the ranking of retrieved results.
* Final Response: The output of the re-ranker is returned as the final response.
{#ais-rag__ul_zc4_3b3_ggc}  
The following diagram shows how AI Search RAG components work together.

The AI Search RAG system works by first organizing data into a searchable format using specific indexing configurations. When a user asks a question, the system processes it and may rephrase the query for better
understanding. It then searches for relevant information with the help of retriever and the re-ranker uses AI to arrange them in the most useful order. All of these steps work together under the guidance of a search profile to
produce an accurate and context-rich answer for the user.

## Embedding model in AI Search RAG {#ais-rag__section_ndn_2zk_zfc}

AI Search RAG uses the advanced search methods, such as semantic or vector search to find relevant and more context-oriented information from the configured indexed sources. The semantic search further uses
embedding model to retrieve the most relevant data based on the user search query for embedding generation. This retrieved information along with the search query then prompts the LLM model to generate the relevant response.

An embedding model is the engine behind RAG that helps it to search and retrieve the correct information from the indexed sources and embed that information into vector map before sending it to the LLM model. ServiceNow
AI Search RAG uses the default ServiceNow Embedding (E5) model and it also supports additional third-party embedding models, such as Azure OpenAI Embedding and Gemini Text Embedding. You can also bring your own third-party hosted custom
embedding model to create embeddings for your RAG use case.

## Activating AI Search RAG (Retrieval-Augmented Generation) {#ais-rag__section_fl2_rhl_zfc}

AI Search RAG (Retrieval-Augmented Generation) functionality is provided by the AI Search RAG plugin (sn_ais_rag). This plugin is automatically activated for your instance when you install [Generative AI Controller](https://www.servicenow.com/docs/access?context=installing-generative-ai-controller&version=zurich&pubname=zurich-intelligent-experiences&ft:locale=en-US) or any [ServiceNow Otto® application](https://www.servicenow.com/docs/access?context=platform-now-assist-landing&version=zurich&pubname=zurich-intelligent-experiences&ft:locale=en-US).