Reviewing Individual Conversations
Read what users actually asked - and use it to make your agent better.
Table of Contents
A customer emails you. They say your agent told them the wrong cancellation policy and they were charged for another month. You know the policy is in your knowledge base. You need to know exactly what the agent said, why it said it, and whether it is saying the same thing to other people.
Conversation and Training is where you go. Every exchange your agent has ever had is stored here - searchable by keyword, filterable by status, readable in full. Within two minutes you have found the conversation, read the exchange, and understood the failure. Now you know exactly what to fix and whether it is a sources problem, an instructions problem, or a one-off.
In this article, you'll learn:
- How to navigate the Conversation & Training tab
- How to find specific conversations using search and filters
- How to read a conversation thread
- What to do with what you find
Opening Conversation & Training
- Go to the Analytics tab.
- Click the Conversation & Training tab.
The left panel lists all conversations. Click any conversation to view its full thread in the right panel.

Searching for Conversations
Use the Search for conversations... bar at the top of the left panel to find conversations by keyword.
This is useful when a specific user reports an issue and you want to find their conversation quickly, or when you want to find all conversations that mentioned a specific topic or product feature.
Filtering Conversations
Three filter controls let you narrow down the list:
Filter by Topic
Click Topics to filter by conversation category. Topics group conversations into common themes automatically - useful for focusing on a specific area of your product or support.
Filter by Contact
Click Contacts to filter by a specific user. The dropdown shows a searchable list of everyone who has started a conversation with your agent.
Use this when a specific user reports an issue and you want to read their exact conversations.
Filter by Status
Click Filter to narrow by conversation status.
| Status | What it shows |
|---|---|
| All | Every conversation - the default view |
| Unresolved | Conversations the agent didn't successfully resolve |
| Resolved | Conversations the agent handled successfully |
| Read | Conversations you've already opened and reviewed |
| Unread | Conversations you haven't reviewed yet |
| Rated | Conversations where the user left a rating |
| Not Rated | Conversations with no user rating |
Reading a Conversation
Click any conversation in the left panel to open the full thread in the right panel. You'll see:
- Every message the user sent
- Every response your agent gave
- Any actions that were triggered
- Whether the conversation was resolved or handed off
Read conversations the same way you'd read a support ticket - look for moments where the agent gave a wrong answer, a vague answer, or no answer at all.
What to Do With What You Find
This is where Conversation & Training becomes genuinely useful - not just as a log, but as a feedback loop for improving your agent.
When the agent gave a wrong answer
Go to Sources and check whether the correct information exists in your connected content. If it does, the source may need to be resynced or the content updated. If it doesn't exist anywhere, add a Snippet with the correct answer.
When the agent gave a vague or incomplete answer
This usually means your sources have the right general topic but not enough detail. Update the relevant article or document in your source, then resync.
When the agent couldn't answer at all
Check Content Gaps in the Overview - this conversation likely shows up there. Add a source or snippet that covers the topic.
When the agent's tone felt off
This is an instructions issue, not a sources issue. Go to Settings → Custom Instructions and add guidance for how your agent should handle that type of question.
When a conversation was handed off unnecessarily
Review your handoff trigger instructions in the Handoffs tab. If the trigger fired for a question your agent should have been able to answer, tighten the trigger conditions.
When a user left a low rating
Filter by Rated and sort for low scores. Read those conversations carefully - low ratings are the clearest signal that something went wrong and the user noticed.
Best Practices
- Make reviewing conversations a weekly habit, especially in the first month after launch - the patterns you spot early lead to the biggest improvements.
- Filter by Unresolved every week as a standing review. These are your agent's failures - reading them reveals what to fix next.
- Don't just look for problems. Filter by Resolved and read good conversations too - understanding what your agent does well helps you replicate those conditions elsewhere.
- When you find a recurring issue across multiple conversations, prioritize fixing it over one-off problems - the impact is much higher.