--- sourceDocument: Australia Platform Analytics sourceDocumentLink: https://www.servicenow.com/docs/r/australia/now-intelligence Release : - australia ft:locale : - en-US ft:publication_title : - Australia Platform Analytics ft:clusterId : - par bundleId : - par workflow : - Platform --- # Guidelines for segments # Guidelines for segments {#ariaid-title1} Release version: Australia Updated March 25, 2026 ![](https://www.servicenow.com/docs/portal-asset/ico-clock) 3 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 Guidelines for segments This document provides best practices for using segments within the semantic layer of ServiceNow's Query Generation. Segments help improve natural language query matching by defining domain-specific filters and concepts that the system can recognize and prioritize. Show full answer Show less ## General Guidelines for Segments * Test segments by posing natural language questions and verify they appear in the Query Generation logs. * Monitor query logs to assess segment effectiveness and identify valuable segments. * Disable noisy auto-generated segments instead of creating many overlapping manual ones. * Use natural, user-friendly phrasing that reflects actual user language. * Maintain one segment per concept to avoid confusion. * Refine segment names and descriptions based on user feedback and observed usage. * Create manual segments for common, recurring questions not well addressed by automated segments. ## Guidelines for Manual Segments * Manual segments are preferred for shipping domain-specific saved searches and receive priority in search results. * Focus on high-value, frequently queried tables such as incident or your primary application tables for maximum impact. * Translate technical filters on choice fields into clear, business-language segment names for better matching (e.g., use "Resolved Critical and High Incidents" instead of filter codes). * Use descriptions to clarify differences between similar segments, helping the language model select the right one. * Keep segment filters simple and focused; break complex use cases into multiple segments to avoid truncation and improve reasoning. * Use manual segments alongside semantic descriptions for tables and columns to provide the language model with both structural and filtering context. * Before creating manual segments, check if an automated segment already exists and consider improving its naming instead of duplicating. * After deployment, monitor segment matches and query correctness, iterating on segment names and descriptions as needed. ## System Property Tuning Suggestions * **Segments not matching user questions:** Lower `segmentsmatchthreshold` (e.g., to 0.60) to accommodate domain-specific terminology. * **Manual segments overshadowed by automated ones:** Increase `manualsegmentscalefactor` to boost manual segments, but only after ensuring segment names and descriptions are well-crafted. * **Too many irrelevant segments in prompts:** Lower `segmentsresultlimit` to reduce noise or raise `segmentsmatchthreshold` to improve match quality. * **Filters truncated due to length:** Increase `maxfilterlength` or split complex filters into multiple segments. * **Batch search disabled:** Enable the `segments.aisbatchfetchenabled` property to improve manual segment coverage by separating manual and automated segment searches. ## Practical Impact for ServiceNow Customers By following these guidelines, ServiceNow customers can ensure their natural language queries are accurately matched to relevant data segments, improving search relevance and user satisfaction. Manual segments allow tailoring of domain-specific saved searches with higher priority, enabling better handling of frequent and critical queries. Monitoring and tuning system properties helps optimize segment performance and result quality in your specific environment. Follow these suggestions to help you use segments in the semantic layer effectively. ## General guidelines for segments {#general-guidelines-segments__section_um4_mpd_y3c} * Test segments by asking natural language questions that should match them. The segment should appear in Query GenerationLogs. * Monitor query logs to verify segments are being matched correctly and to identify the most valuable segments. * Disable noisy auto-generated segments rather than trying to overfit with multiple manual ones. * Use natural phrasing that matches how your users actually speak. * Keep one segment per concept to avoid confusion. * Refine segment names and descriptions based on user feedback and usage patterns * Consider creating manual segments for recurring questions that are not handled well by the existing segments. {#general-guidelines-segments__segment-best-practices-list} ## Guidelines for manual segments {#general-guidelines-segments__section_bg3_qpd_y3c} Manual segments are the recommended way to ship domain-specific saved searches with your application. They receive a priority boost over automated segments during search, and the LLM is instructed to retain all their filters unless completely irrelevant.{#general-guidelines-segments__manual-segments-intro} Start with high-value, high-traffic tables : Focus manual segments on the tables your users ask about most. A handful of well-crafted segments on incident, sc_req_item, or your application's primary table will have more impact than broad coverage across rarely-queried tables. Pair every choice-heavy filter with a clear name : If your segment filters on choice values---for example, `state=6^priority<=2`---the segment name must translate those codes into business language. The LLM matches on the name, not the filter. "Resolved Critical and High Incidents" is searchable, but `state=6^priority<=2` is not. Use descriptions to disambiguate similar segments : If you ship multiple segments for the same table---for example, "Open Incidents" and "Open Critical Incidents"---descriptions are what help the LLM pick the right one. Without descriptions, the LLM may arbitrarily choose between two close matches. Keep filters focused : A segment with 15 filter conditions is hard for the LLM to reason about and likely to be truncated. If you have a complex use case, split it into multiple segments with simpler filters. The LLM can combine filters from different segments when constructing a query. Complement semantic descriptions with segments : Manual segments and semantic descriptions (entity/column config) work together. Descriptions help the LLM understand *what* a table and its columns represent. Segments help the LLM understand *how users typically filter* that table. For best results, ship both. Avoid duplicating automated segments : Before creating a manual segment, check whether an automated segment already covers the same filter. If one does but has a poor name, consider whether improving the source---for example, renaming the report---is a better approach than creating a duplicate manual segment. Review after deployment : After shipping, monitor which segments are being matched and whether the generated queries are correct. If a segment is matching but producing wrong results, the issue is usually the name being too generic or the filter being too broad. Iterate on the name and description first before adjusting system properties. {#general-guidelines-segments__manual-segment-recommendations} ## Property tuning suggestions {#general-guidelines-segments__section_w43_xm2_y3c} Use these suggestions to adjust system properties when segments are not performing as expected. For a complete list of all segment properties and their defaults, see [Query Generation properties](https://www.servicenow.com/docs/zkRx3uAT0Sa4Zf7lTp4Gww "The properties included with Query Generation apply to segments. They define whether the record or indicator that is the basis for a segment has been recently used or changed.").{#general-guidelines-segments__tuning-intro} Segments are not matching user questions : Lower segments_match_threshold (For example, set it to `0.60`). The default `0.70` can be too strict for domain-specific terminology where the user's phrasing differs from the segment name. Start lower and tighten once you have enough segments. Manual segments are being overshadowed by automated ones : Increasing manual_segment_scale_factor gives manual segments a stronger edge, but a higher boost does not necessarily mean better results. A poorly named manual segment boosted above a genuinely relevant automated segment will degrade query quality. Before adjusting the boost, first check whether the manual segment's name and description are well-written. That is usually the real fix. Only increase the factor if you're confident your manual segments are high quality and still losing to weaker automated matches. Too many irrelevant segments in the LLM prompt : Lower segments_result_limit (for example, `6` or `8`). Fewer segments in the prompt means less noise for the LLM, but also less coverage. Find the balance for your domain. Alternatively, raise segments_match_threshold if you're seeing low-quality matches slip through. Long filters getting truncated : Increase max_filter_length if your segments require complex filters. Be aware that longer filters consume more of the LLM's context window and may reduce overall generation quality. Consider splitting a complex segment into multiple focused segments instead. Batch search is disabled and manual segments are not surfacing : Enable the AI Search property segments.ais_batch_fetch_enabled. Without batch search, manual and automated segments compete for the same result slots. Batch search gives each type a dedicated search, significantly improving manual segment coverage. {#general-guidelines-segments__tuning-suggestions-list} **Related concepts** * [Manual segment data model and sync behavior](https://www.servicenow.com/docs/wi4RahSzmKRMDbOZszvYLA "Manual segments use a two-table data model with automatic synchronization between the configuration table and the runtime table used for search operations.") * [Shipping manual segments via plugins](https://www.servicenow.com/docs/LVUOH6Um2VqZu_SlsoWs6w "Business unit application developers can ship manual segments with their applications to provide domain-specific saved searches that work from the moment the app is installed.") **Related tasks** * [Create a manual segment](https://www.servicenow.com/docs/aPVzsc7wyGsf2jz7ThwDlA "Manual segments are admin-created saved searches with friendly names that bridge natural language questions and database filters for the Query Generation semantic layer.") *[\>]: and then