Now Assist Virtual Agent(NAVA)- Configuration Issues Affecting Generative AI Responses<!-- /*NS Branding Styles*/ --> .ns-kb-css-body-editor-container { p { font-size: 12pt; font-family: Lato; color: var(--now-color--text-primary, #000000); } span { font-size: 12pt; font-family: Lato; color: var(--now-color--text-primary, #000000); } h2 { font-size: 24pt; font-family: Lato; color: var(--now-color--text-primary, black); } h3 { font-size: 18pt; font-family: Lato; color: var(--now-color--text-primary, black); } h4 { font-size: 14pt; font-family: Lato; color: var(--now-color--text-primary, black); } a { font-size: 12pt; font-family: Lato; color: var(--now-color--link-primary, #00718F); } a:hover { font-size: 12pt; color: var(--now-color--link-primary, #024F69); } a:target { font-size: 12pt; color: var(--now-color--link-primary, #032D42); } a:visited { font-size: 12pt; color: var(--now-color--link-primary, #00718f); } ul { font-size: 12pt; font-family: Lato; } li { font-size: 12pt; font-family: Lato; } img { display: ; max-width: ; width: ; height: ; } } 1. Outdated or mismatched plugins break the generative AI pipeline What happens: You activated Now Assist in Virtual Agent but generative AI responses aren’t loading, or you’re seeing “Sorry, there was a problem on my side trying to complete the request” errors. Everything looks configured but the assistant either stalls or returns a generic error message. What can get overlooked: Now Assist in Virtual Agent depends on a chain of plugins, and the parent-child relationship means updates need to start with the parent plugin. Often only one child plugin gets updated independently, which can result in version mismatches. The same error message can also appear when the TextToResult capability defaults to a model that isn’t available in your environment (see Scenario 2). Now Assist in Virtual Agent runs on these important plugins. Update the parent and the rest follow: sn_nowassist_vasn_ais_assistsn_ais_ragsn_nowassist_adminsn_vad_genaisn_nava_configsn_generative_aisn_nowassist_gssn_skill_buildersn_ai_websearchsn_aiasn_kg Try this instead: Go to System Definition > Plugins and start your update from the parent plugin (for example, “Now Assist for ITSM”) rather than updating individual child plugins. This ensures all dependent plugins update together and stay version aligned. If you’re seeing errors after a partial plugin update, repair the Generative AI Controller plugin and the Knowledge Graph plugin as a first step. This has resolved the issue in several reported cases. See: How to update Now Assist plugins after a patch See: Resolve search errors after setting up enhanced chat in Now Assist in Virtual Agent 2. The TextToResult capability defaults to an unavailable model What happens: When testing, you get “Sorry, there was a problem on my side trying to complete the request.” The plugins are up to date and the configuration looks correct, but the error persists. If you’ve already ruled out plugin version mismatches (Scenario 1), this is the next place to check. What can get overlooked: The TextToResult capability in the OneExtend configuration can default to a model that isn’t available in your environment. For example, the default may point to “TextToResult (GPTLarge)” when your instance uses NowLLM. There's no indicator at steup time. The mismatch only becomes visible when the conversation runs. How to check: In the navigation filter, enter sys_one_extend_capability.list and search for TextToResult. Open the record and review the OneExtend Definition Configs related list. Check which configuration is set as the default. Try this instead: If your environment uses NowLLM, disable the default for “TextToResult (GPTLarge)” first, and then switch the default to “TextToResult (NowLLM).” The order matters: Disable the current default before enabling the new one. After making the change, clear the browser cache, close all windows, and open a new chat from the designer. Test using both standard and enhanced chat experiences. See: Resolve search errors after setting up enhanced chat in Now Assist in Virtual Agent 3. Application scope access restrictions block Now Assist from reaching resources in other scopes What happens: You type something in the Virtual Agent conversation window and it gets stuck at “Figuring out next steps” with no error message, just an indefinite wait. What can get overlooked: This can happen in environments where Now Assist needs to access resources in another application scope, for example, HR. ServiceNow uses application scope access records to control which scopes can call into other scopes. If the Status field on the access record between the AI Agents scope and the target scope is set to Restricted, the call fails silently. There’s no error in the conversation. How to check: Go to System Applications > Application Cross-Scope Access and review the records for cross-scope calls between the AI Agents application scope (sn_aia) and the scope where the target resources reside (for example, HR). Try this instead: Change the Status field on the access record from Restricted to Allowed so that cross-scope calls are permitted from the AI Agents scope to the target scope. See: How to resolve Now Assist not responding in Virtual Agent in Service Portal 4. The feedback topic only appears in Now Assist chat experiences What happens: You’re trying to configure the feedback topic for post-conversation surveys, but the topic won't appear. This is because topics are scoped to Now Assist chat experiences by design. How to check: Go to Conversational Interfaces > Assistants, open your assistant, and check the chat experience. The feedback topic appears only within Now Assist chat experiences. Try this instead: Go to Conversational Interfaces > Assistants (not Settings > Virtual Agent > Custom greetings and setup). Open the assistant and select Now Assist in Virtual Agent. The feedback topic should appear in the survey topic list. Make sure the topic is published. Unpublished topics won’t appear regardless of the experience context. See: Now Assist - Virtual Agent Feedback topic is missing from the Topics list 5. AI Search and key properties are not configured for generative responses What happens: The conversation flows work but the answers aren’t useful. Responses are poor quality, knowledge articles don’t surface, or the generative AI answers are inconsistent. What can get overlooked: Response quality depends on several properties and configurations that aren’t part of the standard activation flow. No single setup screen brings all of them together, so it’s possible to miss one or two and not realize they’re affecting response quality. The three properties that most commonly cause response quality issues: AI Search must be enabled. Verify the status at AI Search Admin > AI Search Status. If AI Search is not active, generative responses have no knowledge source to draw from. The auto proxy property must be set to true. The property sn_vad_genai.com.glide.cs.one_extend.auto_proxy_enabled enables the Java-based execution path, which is faster and more reliable than the Flow Designer execution path. If this is set to false, responses may be slower or fail intermittently. Document ingestion must be complete. The property sn_ais_assist.dpr_ingestion_completed should be set to true. If documents haven’t been fully ingested, search results won’t return in the generative responses. Try this instead: Before testing response quality, verify all three properties are set correctly. Then confirm the following settings on the Now Assistant configuration: Skills are enabled and search sources point to the correct knowledge basesGenius result is set to Multi ContentDisplay experience is set to the expected configurationChat experience is set to Enhanced (for the new experience) or Standard (for the older experience) Once all settings are confirmed, test using the assistant’s built-in Test Assistant function (Conversational Interfaces > Assistants > choose the assistant > Review > Test your Assistant). The preview mode gives you access to the complete logs tab, which helps diagnose response quality issues. See: How to author knowledge articles optimized for Now Assist in Virtual Agent 6. Diagnostic data is spread across multiple tables with no obvious starting point What happens: Something goes wrong in a conversation and the diagnostic data exists, but it’s spread across several system tables. Without knowing the right sequence, troubleshooting takes longer than it should. What can get overlooked: Now Assist generates diagnostic data across several tables, and each table serves a different purpose. Knowing which table to open first, and what to look for in each, turns a long hunt into a short diagnosis. The diagnostic path: Start with sys_cs_conversation. Pull the “calling topic” column on the conversation task related list. Sort descending, then review bottom-up. This reveals the sequential flow of the conversation and where it broke.Check sys_cs_fdih_invocation. Review the input to see which skills were applicable (match these against sys_gen_ai_skill). Review the output to check whether the geniusResultsTemplate object contains the expected data. Also check the timeout field. Timeouts are a common cause of silent failures.Check sys_gen_ai_log_metadata. Filter by created by, created date, and meta document fields for the specific request. Review the Error and Error Code fields. If the error code isn't self-explanatory, match it against the generative AI error code definitions. See: Generative AI error code definitions