What Can Be Tuned in ServiceNow Otto for Virtual Agent (Now Assist in Virtual Agent)Issue Administrators commonly ask what can be adjusted beyond selecting search sources and a model provider. This article summarizes the documented configuration surfaces for tone, response format, persona, search and citation behavior, and model provider selection, and states plainly which commonly-requested settings are not documented. Release Australia Resolution Tone, response length, and persona Each assistant can be branded independently through its Personalization settings in Assistant Designer. Three controls are available per assistant: Tone — Direct (default), Empathetic clear, or Professional informative.Response length — Succinct (default), Balanced, or Detailed. This is a guide rather than a strict limit: results vary with the complexity of the user's query, so a short factual question may still receive a short answer regardless of the setting chosen.Persona — a nickname for the assistant and a short description of its subject-matter specialty, used to give the assistant a distinct voice in conversation. These settings apply per assistant, not instance-wide, so different assistants (for example, an HR assistant and an IT assistant) can each use a different tone, response length, and persona. The preview pane shown while configuring these settings is illustrative only — it does not make a live call to the underlying model — so it should not be relied on to predict the exact wording of real answers. On earlier releases, the Personalization section could be hidden behind a system property that administrators needed to enable. Starting with the Australia release, that property was removed and Personalization is shown by default. If Personalization is not visible on an earlier release, check whether that property is present and correctly configured. Search sources and citation behavior Search sources (knowledge bases, conversational catalog items, external content feeds, and so on) are assigned per assistant. Each assigned source has an "Include AI Responses" setting, and only sources with this setting enabled actually feed the large language model that produces the assistant's synthesized conversational answer. Sources without it enabled may still be searched and listed, but they do not contribute text to the generated answer. Separately, AI Search (the search engine behind Now Assist in Virtual Agent's search and citation results) offers its own tuning capabilities that influence search relevance generally: Result improvement rules — boost, block, or promote specific results using weighted rules.Synonym dictionaries — link a synonym dictionary to a search profile so equivalent terms match each other.Knowledge Graph schema personalization — enabled by default when AI Search is active on an assistant; this affects which Knowledge Graph schema content is included in search.Machine learning relevancy — AI Search automatically tunes relevancy over time based on user interactions; this happens automatically and does not require manual configuration. These are documented as AI Search capabilities that influence search results and ranking. ServiceNow documentation does not state that any of them change how the assistant's large language model synthesizes or composes its conversational answer — they should be treated as search-tuning tools rather than a way to directly steer the generated answer text. The wording of the citation message (the "Sources" text shown to the end user alongside an answer) can be customized separately. This changes only the message text itself — it does not change which sources are cited or how closely they match the query. Model provider The underlying AI model provider is changed from AI Admin Hub > Settings > Manage AI models > Manage model providers. The provider can be set at three levels of scope: Instance — applies to all skills that don't have a more specific override.Skill group — applies to a related group of skills. Now Assist in Virtual Agent falls under the "AI agents" skill group.Individual skill — applies to one specific skill only, overriding the instance and skill group settings. This allows an administrator to run Now Assist in Virtual Agent on a different model provider than the rest of the instance, or to pin a specific skill to a specific provider, without making an instance-wide change. What is not configurable As of the Australia release, no documented setting exists for any of the following without customization: Retrieval depth (how many candidate results are considered before an answer is generated)Similarity or confidence thresholds for what counts as a usable matchRe-ranking of candidate search resultsThe number of citations shown per answerFree-form editing of the assistant's underlying system prompt Related Links Brand and personalize an assistantEdit a chat assistantAssign search sources to a chat assistantAdd a Knowledge Graph schema to a chat assistantCustomize citation messageSummary Genius ResultsKnowledge base articles Genius ResultsResult improvement rulesBoost search results using a result improvement ruleMachine learning relevancy in AI SearchLink a synonym dictionary to a search profileImproving the user experience with AI SearchManage model providers