Theme
Knowledge Search
In addition to automatic suggestions triggered by customer utterances, you can search the knowledge base manually at any time using the search bar at the top of the Assist tab.
When to Use Manual Search
Manual search is useful when:
- The automatic suggestion did not fully address the customer's question.
- You want to explore a related topic before the customer asks about it.
- The customer's question was ambiguous and you want to search with clearer phrasing.
- You need to look up a specific policy, procedure, or product detail.
Using the Search Bar
Knowledge search bar at top of the Assist panel with search results displayed below
- Click the search bar at the top of the Assist tab.
- Type your question in natural language.
- Press Enter or click the arrow button to submit the search.
While the search is processing, the button turns red with a spinning icon. Results appear as suggestion cards in the suggestions area, identical in format to auto-generated suggestions -- each includes a streaming answer, source citations, relevance scores, and feedback controls.
Auto-Suggestions vs. Manual Search
Agent Assist generates suggestions in two ways:
| Trigger | How It Works | When It Fires |
|---|---|---|
| Automatic | Customer speech or messages are captured in real time and sent to the RAG pipeline. Suggestions appear without any action from you. | Voice: as the customer speaks. Digital: when a customer message arrives. |
| Manual | You type a question in the search bar. | When you press Enter. |
Both types produce the same kind of suggestion cards. They work together -- auto-suggestions keep flowing while you can run manual searches at any time.
TIP
If the knowledge_search feature is disabled by your administrator, the search bar will not appear. You will still receive auto-generated suggestions.
Participant Role Context
When the system processes a search, it considers the participant role to tailor results:
- Customer utterances (auto-triggered) are treated as questions from someone unfamiliar with internal processes. The system prioritizes customer-facing content.
- Agent searches (manual) are treated as queries from a knowledgeable user. The system may surface more detailed, internal-facing content when available.
This distinction helps the RAG pipeline return the most appropriate level of detail for each context.
Tips for Effective Searches
TIP
Write your search queries as specific, complete questions for the best results.
Be specific. Instead of searching for "refund," try "what is the refund policy for subscriptions canceled after 30 days." Specific queries return more targeted results.
Use customer language. If the customer used particular terminology, include it in your search. The knowledge base may have content indexed with those exact terms.
Ask one question at a time. The RAG pipeline works best with a single, focused question. If you need answers to multiple questions, run separate searches.
Rephrase if needed. If the first search returns low-relevance results, try rephrasing. Use synonyms, change the sentence structure, or add more context.
Check the sources. After a manual search, review the source citations to verify the answer is grounded in your organization's content.
Next Steps
- Learn about quick reply chips and follow-up questions: Quick Replies & Follow-Up Questions
- Understand suggestion cards in depth: Suggestion Cards