What AI Search changes
Traditional ServiceNow search uses keyword matching — it finds articles and records containing the exact words you searched for. AI Search uses semantic similarity — it understands the intent of your query and matches it against the meaning of documents, even when the exact words do not appear.
Example: a user searches "my laptop won't turn on." Traditional search finds articles containing "won't turn on." AI Search understands this is a power/boot issue and returns articles about battery replacement, power adapter troubleshooting, BIOS settings, and hardware failure — even if they never use the phrase "won't turn on."
How AI Search works technically
AI Search converts documents to vector embeddings — mathematical representations of a document's semantic meaning. When a user searches, their query is also converted to an embedding, and the system finds documents whose embeddings are mathematically close to the query embedding. This is why semantically similar content matches even without exact keyword overlap.
Setup requirements
- ServiceNow Washington DC release or later
- Unified Search plugin (com.sn_unified_search) activated
- AI Search entitlement — verify with your account team
- Knowledge base articles published and indexed
Configuring AI Search
- Navigate to System Definition > AI Search Administration
- Create an AI Search Profile — defines which tables are searched and which fields are included
- Add Sources — which knowledge bases, catalog items, and tables are searchable
- Configure Field Weights — which fields carry more weight in relevance ranking (title usually weighted higher than body)
- Run initial indexing — AI Search builds embeddings for all existing documents on first setup
Embedding generation and indexing
The initial indexing step converts all existing documents to vector embeddings. For large knowledge bases, this takes hours on first run. After initial indexing, new and updated documents are indexed incrementally — there is typically a short delay (minutes) between publishing an article and it appearing in AI Search results.
Testing search quality
Use the Search Admin Test panel to query your index and see: which results are returned, their relevance scores, and which source they came from. This is the tool for identifying gaps in knowledge base coverage — queries that return no results or irrelevant results indicate articles that need to be written or updated.
Related guides: Now Assist overview · Activation guide
How AI Search differs from keyword search
Traditional ServiceNow search matches words. A user searching "laptop won't turn on" only finds knowledge articles that contain those exact words. AI Search uses semantic embeddings — mathematical representations of meaning — to find content that addresses the user's intent even when the words are different. "Laptop won't turn on" finds articles about power supply failures, battery issues, and hardware diagnostics even if none of them use the phrase "won't turn on."
The practical result: knowledge deflection rates improve because users find relevant articles they would have previously missed, and fewer contacts reach the service desk for issues that existing knowledge articles already address. ServiceNow's published benchmarks show 20–40% improvement in self-service resolution rates after AI Search activation, with the biggest gains in organisations whose knowledge base is comprehensive but poorly tagged.
Knowledge base quality requirements
AI Search amplifies your existing knowledge base quality — it does not fix a bad one. Before activating, audit your knowledge base: are articles well-written with clear titles that describe the problem? Are articles up to date — no articles referencing legacy systems or outdated procedures? Are articles appropriately scoped — one article per issue, not one massive article that covers ten unrelated topics? AI Search reads and indexes article content; articles with vague titles or poor structure produce lower-quality search results regardless of the AI layer above them.
Enabling AI Search
Activate the com.sn_now_assist_search plugin (requires base Now Assist Platform plugin active first — see activation guide). After activation, AI Search indexes your knowledge base. Initial indexing takes time proportional to knowledge base size — a large knowledge base with thousands of articles may take several hours for full initial indexing. Monitor indexing progress in the AI Search administration dashboard. Search quality improves as the index builds — do not judge results in the first hour after activation.
Related: Now Assist overview · Now Assist for ITSM · Activation guide · Implementation strategy
AI Search and knowledge gap identification
One underused capability of AI Search is its ability to identify gaps in your knowledge base. The AI Search analytics dashboard shows search queries that returned results vs those that did not — the zero-result searches reveal topics users are looking for that your knowledge base does not cover. This is actionable data for your knowledge management programme: the zero-result queries are a prioritised list of knowledge articles to create.
Low-click-through searches (searches that returned results but the user did not click any article) indicate mismatched articles — the results exist but did not match what the user was looking for. Review these periodically and update or retag the articles to better match common search intent. AI Search quality is not a set-and-forget configuration — it improves with ongoing knowledge base curation and periodic review of the search analytics. Assign someone to review the analytics dashboard monthly as part of your knowledge management process.
AI Search configuration beyond activation
After initial activation and indexing, several configuration options improve AI Search quality for your specific environment. Synonym configuration: navigate to AI Search > Synonyms and add organisation-specific synonyms (your internal application names, acronyms, product names) that users search for but that may not appear verbatim in knowledge articles. Stopword configuration: words that are very common in your environment and should be ignored in search ranking. Featured results: promote specific knowledge articles for high-volume search queries to ensure the best answer appears first. These configurations are cumulative — each one incrementally improves search relevance for your specific user base and knowledge base vocabulary.
Review AI Search analytics quarterly. Search query volume, click-through rates by article, and zero-result query lists all improve as you act on them. Schedule a 30-minute analytics review with your knowledge management team quarterly and allocate time for the follow-up actions — creating missing articles, updating poorly-performing ones, adding synonyms for newly identified search patterns.
AI Search in the Service Portal
AI Search replaces the standard keyword search in Service Portal automatically after activation and indexing — no portal configuration changes are required for the basic integration. The search bar on your portal switches to semantic search without any widget changes. For portals with custom search implementations (a custom search widget rather than the standard one), additional configuration is needed to route searches through the AI Search API rather than the legacy search. Check your portal's search widget type before activating to determine whether the upgrade is automatic or requires custom widget updates.
Measuring AI Search ROI
Establish these baseline metrics before activating AI Search, then measure at 30 and 90 days: search-to-ticket ratio (what percentage of portal searches end in ticket submission vs self-resolution?), knowledge article view rate per search, and time spent on search before ticket submission. After activation, improvements in all three metrics indicate AI Search is delivering value. The most directly measurable ROI is the reduction in tickets submitted for issues that existing knowledge articles address — each deflected ticket is a measurable cost saving. Calculate your cost per ticket handled by the service desk, multiply by the number of additional deflections attributed to AI Search, and you have a concrete ROI figure to present to leadership.
AI Search and the knowledge lifecycle
AI Search makes the quality of your knowledge lifecycle management more visible. Articles that are outdated, redundant, or poorly written produce worse search results than good articles — the AI cannot compensate for poor source material. Use the activation of AI Search as an opportunity to review and clean up your knowledge base: archive articles that are more than two years old without recent views, merge duplicate articles that address the same issue, and update articles where the described procedure has changed. A smaller, higher-quality knowledge base produces better AI Search results than a large, mixed-quality one. Set a regular cadence for knowledge base maintenance — quarterly is the minimum for environments where systems and procedures change frequently.
AI Search is the Now Assist feature with the longest payback period — it requires knowledge base investment before activation, time to index after activation, and ongoing curation to maintain quality. But the compounding effect of better search on knowledge deflection rates, agent productivity, and user satisfaction makes it one of the highest-ROI ServiceNow investments for organisations with mature knowledge bases. See the implementation guide for the phased rollout approach that sequences AI Search correctly within the broader Now Assist programme.
AI Search requires the Xanadu release or later (see Xanadu release guide) and the Now Assist Platform base plugin active before the search-specific plugin can be activated. The initial indexing time depends on knowledge base size — a knowledge base with 500 articles indexes in minutes, one with 10,000 articles may take several hours. Plan activation during a low-traffic period if your instance serves a large user base that actively uses the portal search.
AI Search deployment is most successful when treated as a knowledge management initiative with a technology component, rather than a technology deployment with knowledge management implications. The organisations that get the most value are those that pair activation with deliberate knowledge base improvement — auditing content quality, filling identified gaps, and establishing ongoing maintenance processes. The AI layer amplifies the quality of what is underneath it; invest in the foundation as much as the feature.