Fallbacks in Now Assist Virtual Agent
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
06-12-2026 05:56 AM
Hi! I'm working on creating a Fallback topic (let's call it F1) which fires when the conversing user expresses dissatisfaction with the search results within Virtual Agent. Currently, I have a custom fallback (F2, view attached image) (VA Settings -> Chat Experience -> Legacy Fallbacks) and I've attached it to an LLM topic. And from my understanding, it fires whenever the user's query yields no results, and the AI is unable to effectively answer.
I created an LLM topic, gave it a description that instructs it to fire whenever the user is unhappy, or expresses dissatisfaction, but when I try it out on VA, such queries almost always trigger the custom fallback (F2), when it should actually be triggering F1, since as far as i know, topic discovery fires first before the AI Search defaults to custom fallbacks. Can anyone that's well-versed in Now Assist Virtual Agent & Assistant Designer help me out with this?
Also, is there any way to insert a feedback type question after search results? Something that immediately presents a "Are you satisfied with the response" after every search response to a user's query, so that if the user answers no, I can trigger my fallback from there?
- Labels:
-
AI Search
-
Fallback
-
Virtual Agent
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
a week ago
The key here is to distinguish between a failed topic match and a user who is unhappy with the search results. These are two different scenarios and may need separate handling.
I'd suggest checking the following:
Open Virtual Agent Designer and verify that your LLM topic is enabled for topic discovery.
Check the AI Search Fallback and Fallback setup topics in your active chat experience.
Test what happens when a user's query returns no matching topic versus when AI Search returns results that don't answer their question.
If you want to collect feedback after every search response, create a custom topic with Yes/No choices and route the conversation based on the response.
If the user selects No, redirect them to a clarification topic or a live agent.
The standard Virtual Agent Feedback topic does not guarantee feedback after every search response, as it is presented based on its configured behavior.
Also, if you're using Now Assist Virtual Agent with Unified Search, verify the fallback behavior for your ServiceNow release rather than assuming the traditional AI Search Fallback configuration applies.
I'd test this in a sub-production instance before changing any out-of-box setup topics.