Search result disparities between AI Search and Now Assist search features
Summarize
Summary of Search result disparities between AI Search and Now Assist search features
ServiceNow’s AI Platform® offers multiple search tools—AI Search, Now Assist Genius Results, and Now Assist in Virtual Agent—that may produce different results for the same search query. This variation is expected because each tool uses distinct search methodologies and architectures to generate answers, impacting relevance and consistency.
Show less
Differences in Search Tools and Their Behavior
- AI Search: Uses a traditional keyword search approach that matches individual terms with source data. It applies linguistic processing like lemma normalization, synonyms, stop words, and typo correction. Results are generally consistent for identical queries unless source data changes or machine learning relevancy adjustments occur, which can alter result ordering over time. AI Search does not involve a large language model (LLM).
- Now Assist Genius Results: Employs a hybrid search combining keyword and semantic vector search to understand user intent and meaning. The top matching results are compiled and sent as a single prompt to an LLM for answer generation. Because LLMs are non-deterministic, the same query can yield different answers even when submitted simultaneously. This leads to more variable results compared to AI Search.
- Now Assist in Virtual Agent: Also uses hybrid search but operates via a conversation-based chat interface with an agentic AI backend. This architecture maintains conversational context and generates responses differently from the Genius Results approach. Therefore, results can vary from both AI Search and Genius Results, even using the same search configuration.
Practical Implications for ServiceNow Customers
- Expect differences in search outcomes across AI Search and Now Assist tools due to their inherent technological differences.
- Using the same search configuration across portals and Virtual Agent chat assistants is recommended to minimize configuration-related discrepancies; however, identical results are not guaranteed because of architectural differences.
- Understanding that AI Search provides more stable, keyword-focused results while Now Assist leverages semantic understanding and LLM-generated answers helps in setting appropriate expectations for end users.
Key Outcomes
- Customers can leverage AI Search for consistent, term-based results suitable for scenarios requiring predictable outputs.
- Now Assist Genius Results and Virtual Agent provide richer, intent-driven answers that may vary but offer more conversational and context-aware interactions.
- Being aware of these differences enables better configuration and user experience planning across ServiceNow search and conversational AI capabilities.
The ServiceNow AI Platform® offers a variety of search tools, which may return different answers for the same or similar searches. This disparity in results is expected. It occurs because each tool uses a different approach and architecture to find results and generate answers that match your search.
In this example, each search tool returns a different answer even though the user's search (or question) is the same in all three tools. This is expected behavior, since the search tools all handle searches differently.
AI Search
The AI Search engine uses keyword search, meaning that it looks for the best matches for your search terms in its indexed source data. Search features such as lemma and Unicode normalization, synonyms, stop words, and typo handling may modify the set of terms that AI Search considers matches for your search, but the matching is always done on a per-term basis.
Keyword search should return consistent results for the same search until your source data is updated. Machine learning relevancy can affect the exact order in which your results appear over time, though, so even if your data doesn’t change, your search may not return exactly the same result set today as it did the previous month.
Now Assist Genius Results
The Now Assist Genius Result configurations offered in Now Assist in AI Search use a hybrid search mode. This mode blends keyword search with semantic vector search to find results in the AI Search index based on the intention and meaning of your search as well as on the best term matches.
Between the non-deterministic LLM behavior and the difference in search matching modes, Now Assist Genius Result answers can be expected to vary significantly from results returned by AI Search for the same search terms, even when using the same search configuration in both tools. This variance is expected because the two search tools take such different approaches when finding and generating answers for your search.
Now Assist in Virtual Agent
Now Assist in Virtual Agent uses hybrid search to find matching results in the AI Search index, like the Now Assist Genius Result configurations do.
Unlike those Genius Result configurations, however, Now Assist in Virtual Agent has its own agentic AI back-end architecture for generating responses. This architecture doesn't rely on sending a single prompt to the LLM for answer creation the way the Now Assist Genius Result architecture does.
As previously described, using the same search configuration in the portal and your chat assistant doesn't guarantee that you will see exactly the same search results in Now Assist in Virtual Agent as you see in the portal using AI Search, but it does remove one possible source of difference.
Summary of differences between search tools
| Search tool | Search mode | Interaction | LLM usage |
|---|---|---|---|
| AI Search | Keyword search | Query-based (search field) | None |
| Now Assist Genius Results | Hybrid (blend of keyword and semantic vector) search | Query-based (search field) | Most relevant search results sent to LLM in a single prompt for answer generation |
| Now Assist in Virtual Agent | Hybrid (blend of keyword and semantic vector) search | Conversation-based (chat) | Agentic AI which maintains conversational context when submitting prompts to LLM |