Standard chat

Now Assist in Virtual Agent standard chat is a conversational support experience within a static chat window.

Now Assist in Virtual Agent enhances the user experience by combining AI Search with generative AI chat skills. These skills can speed up issue resolution and reduce deflection to a live agent.

Note:

For more information about selecting a chat experience in the admin guided setup, see Display your chat assistant on a portal, channel, or mobile app.

Now Assist in Virtual Agent provides your users with an interactive generative AI experience. A friendly, natural language conversation is easier to understand and helps users to be more comfortable with chatting with a bot.

The following sections highlight how a conversation powered by generative AI might look in Virtual Agent.

Starting the conversation

The following figure shows an example of the beginning of a new conversation.

Figure 1. Now Assist greeting

Table 1. Now Assist greeting elements

Element	Description
1: Bot greeting	Note: The Answers generated by AI. Review for accuracy. message prior to the greeting can't be edited or turned off. After greeting the user, the bot provides an explanation of what the user can expect. Because the large language model (LLM) is still learning, the bot indicates that it may make mistakes, and that the user should check answers for accuracy.
2: User request	The user can enter any utterance.
3: Response indicator	While the LLM processes the utterance, an animated avatar appears alongside latency feedback messages in the chat window to let the user know that the bot is working.

Chatting with the Virtual Agent

After the user enters an utterance and a search result is found, a synthesized response may appear. A synthesized response includes a brief summary of the requested information and search results along with Genius Results. For more information on how these search results are found, see Now Assist Actions Genius Results and Now Assist Q&A Genius Results.

If multiple results are found, they are presented to users, and users can view inline citations and/or the Sources section at the bottom of the synthesized response. Users can view Knowledge Base articles, Virtual Agent topics, subflows or actions, external content, people citations, and complete catalog request items. Virtual Agent can only return available catalog items that match a user's request when the Now Assist Multi-Turn Catalog Ordering skill is enabled. Users can follow up on a synthesized response and receive an answer that considers the previous context. For example, if you previously were chatting about the company's travel policy and then enter International, the system recognizes that you’re wanting additional information on the company's international travel policy.

Note:

For full catalog functionality in the chat window, enable the generative AI experience for catalog item request submissions. For more information, see Configure Now Assist in Conversational Catalog Request.

Whenever a single Virtual Agent topic is found within a synthesized response, users are brought directly into that action's flow. You may consider bypassing the synthesized response and options by automatically launching catalog items, too. For more information on automatically launching single search result actions, search for the sn_nowassist_va.synthesized_autostart_items system property in Available system properties.

Responses generated in Now Assist in Virtual Agent can come from a combination of catalog items, Virtual Agent topics, subflows and actions, Knowledge Base articles, attachments, external content, and people sources. The following citations may appear inline and/or within the expandable Sources section depending on what search results are returned:

• Catalog
• Topic, subflows, or actions
• Q&A Knowledgebase articles
• External content connections
• People

If you select an inline citation, a popover appears containing either additional information or an action to start a request.

Figure 2. An inline citation popover example linking to a source or action

Responses can also be generated based on content from the catalog item description and the Virtual Agent topic description. When this occurs, citations lead the user to one of the items in the Here's what you can do next section of the response.

Each Virtual Agent response includes a feedback icons panel. The user can indicate if the response was helpful by selecting the like thumbs up icon (). If the response wasn't helpful, the user can select the dislike thumbs down icon (). This feedback is used to train the LLM model and improve responses over time.

If AI Guardian is enabled and your request contains profane content, the Virtual Agent responds with a message prompting you to re-enter an appropriate request without profanity or offensive content. If your request is too ambiguous in a Virtual Agent chat, Virtual Agent asks a clarifying question and displays a citation with an action. Citations are presented starting from the second clarifying question. Select a citation to start the request. If there are no actions relative to the request, citations aren’t shown.

If you need more help, Now Assist in Virtual Agent directs you to either a live agent, resets the conversation, or creates an incident. The option is determined by the admin in the Now Assist in Virtual Agent guided setup. For more information, see Configuring assistants overview.

1. Live agent support routes the user to an available agent when configured.
2. Reset the conversation restarts the conversation to the beginning to ask another question.
3. Record producer directs the user to a catalog record producer to create an incident or case.

If you entered a single message with multiple questions or requests, a Ready to move on to your next task? card appears. For example, a response like Order a loaner laptop and also what is spam? When you select the Continue to next task button on the Ready to move on to your next task? card, Virtual Agent searches for a response to your second question or request. In this example, Virtual Agent would search for an answer to your what is spam? question.

If an administrator didn’t configure any skills, no available skills are shown. However, the chat bar remains available for the user to enter an utterance. This triggers AI Search for a Knowledge Base article or catalog item.

Figure 3. Example of a chat with no skills activated

For information about configuring a Now Assist skill, see Activate a Now Assist skill.

Agentic conversations

Note:

Admins must first enable AI agents before end users can experience agentic conversations. For more information on how to set up AI agents for agentic conversations, see Agentic conversations in Virtual Agent. Now Assist in Virtual Agent discovers and executes AI agents. For more information about AI agents, see Now Assist AI agents.

When you submit a single question to the Virtual Agent, the agent understands the query and begins a flow. When you submit a message with multiple questions or requests, Virtual Agent answers the multiple questions consecutively in a response. It can reason, plan, and execute across AI agents, Virtual Agent topics, conversational actions and subflows, catalogs, Knowledge Base articles, custom skills, and any Now Assist in Virtual Agent supported skills to help you. Before receiving a response, you receive acknowledgment messages from the Virtual Agent and on-screen messages to let you know where the agent is at in the agentic processing flow. After the processing has completed its flow, a View AI Steps section header appears, where the processing flow steps can be expanded and viewed. You can stop the agentic processing flow at anytime by selecting the End flow icon (). After an action starts, it cannot be stopped. Selecting the End flow icon () only stops the proceeding processing steps.

Figure 4. Example of an agentic conversation starting

Catalog citations

For conversational catalogs, the Start request option appears, and if selected, the catalog request flow launches within the Virtual Agent. For non-conversational catalogs, the Go to request option appears with an external link that if selected, either redirects you to a new browser or pop-up window to order the catalog item.

Figure 5. Example of a synthesized response with conversational and non-conversational catalog options

Figure 6. Example of a conversational catalog response flow

Figure 7. Example an LLM-generated message before auto-executing the first step in the action

When the catalog order is complete, Virtual Agent displays a card for the order.

Figure 8. Example of a summary confirmation for a completed catalog request

Topic, subflows, or actions citations

The [Asset's name] option appears, and if selected, the asset flow begins in Virtual Agent.

Q&A Knowledge Base citations

The answer is shown with the option to view your sources through the numbered inline citations icon (). These Knowledge Base inline citations show which ServiceNow knowledge articles were used to generate the response.

Note:

If you have activated Knowledge Graph, you may get more personalized Q&A Knowledge Base responses depending on whether the user schema attributes were triggered. For more information about receiving these personalized responses, see Access Knowledge Graph Schema.

External content connections citations

If external search sources are added to your information sources when Configuring assistants overview, you can search through external content connections. This can include information from attachments and knowledge blocks, or from the following list of content connection:

• Microsoft SharePoint
• Confluence
• Atlassian Jira Cloud
• Google Drive
• Microsoft Teams
• Predefined web sources
• ServiceNow® documentation
• Slack

For more information about external content connections, see External Content Connectors.

Figure 9. Example of an inline citation with external content source popover

People citations

Information about people in your organization can be found if you have activated Knowledge Graph. Information about the person you're searching for appears in the synthesized response, along with an in-line numbered citation and people citation in the Sources section. If a direct match doesn't occur and multiple people are found, try providing more information to narrow down the people search results. Up to three people can appear in a synthesized response. You may be asked a clarifying question, such as Are you asking about [user_1], the programmer, or [user_2], the faculty member? to yield better results. Selecting the person's name in the Sources section redirects you to the portal's user profile landing page.

Figure 10. Example of a synthesized response with people results

Selecting the person's inline numbered citation presents a popover. The information in the popover can include the following information:

Note:

The following popover details can be customized through EVAM. For more information about customizing the data details, see Define an EVAM view template.

• Manager
• Location
• Email
• Teams
• Phone
• Shared files
  Important:

  Shared Microsoft SharePoint files between you and the person found appear only on the people popover. These shared files only appear after you have completed the prompt to Log in, signed in successfully, and have entered a new search query for that person. A new search query is necessary after you have signed in so that up to three shared Microsoft SharePoint files can appear. You’re automatically logged out of your Microsoft account every three months and must repeat the Log in prompt again to see shared files.

Figure 11. Example of a people citation's popover

Although rare, when you receive a synthesized response that includes a person, you may also receive other citations associated with that person. For example, a Knowledge Base article that was authored by the person you searched for may also appear.

Extended entities and view records

Extended entity information can be found if you have activated Knowledge Graph.

Note:

To enable the Knowledge Graph natural language query (NLQ) schema, configure this schema for the assistant. To configure the schema for an assistant, see Assign search sources to a chat assistant.

These extended entities that come from the additional custom tables associated with the Knowledge Graph natural language query (NLQ) schema can include:

• Assets
• Incidents
• Recently viewed knowledge base articles
• Requests
• Tasks

Select an entity in-line citation to view that entity record in a popover, or select View records to view a list of additional entities. Selecting a link from the View records pop-up opens a new tab with that entity table's data.

Figure 12. Example of extended entities and view records

The following example is of a conversation showing 3 out of 12 requests and the option to view records and sources.

Fallback options

Note:

For more information about where and how to enable fallback options, see Manage an assistant chat experience.

A fallback state can occur whenever search results are unavailable. Scenarios where search results are unavailable include the Virtual Agent didn't understand the query, complaint small talk was found, or an error occurred. When search results are unavailable, a single or multiple fallback options may appear. These fallback options include:

• Request a live chat: Triggers live agent mode and routes you to a human support representative.
• Search the web: Triggers web search mode and uses the internet to search for the results.
  Note:

  Only the last query entered into the conversation is considered when entering web search mode via this Search the web fallback option.
• Create a generic ticket: Creates a record.
• End this chat: Ends the chat.
  Note:

  This option is only available to standard chat conversations.
• Custom fallback option: Presents a fallback Virtual Agent topic.

Select a fallback option to launch that mode or action.

Figure 13. Example of multiple fallback options