Open the conversation review page

The conversation review page is where you monitor and assess every conversation your AI agent has - whether it's a real customer chat, part of an evaluation run, or a test you started yourself.

When you open the page, you'll see a list of conversations on the left and the selected conversation's history on the right. Three tabs sit above the list:

• Real - conversations between your customers and the agent, happening on your support website.

• Evaluations - the results of evaluation runs, where a set of questions is given to the agent as a dataset to assess its performance.

• Test - conversations you generate yourself from the test sidebar while configuring the agent.

Switching tabs keeps each conversation type separate, so you're never mixing real customer chats with test runs.

Review real conversations

The Real tab shows conversations your customers are having with the agent through the chat widget on your support website.

Select any conversation in the list to read the full exchange on the right. From here you can see how the agent responded, whether it resolved the customer's question, and where a response could be improved.

Review evaluation runs

The Evaluations tab brings your evaluation runs into the same review interface, so you no longer need to open them from a separate page.

An evaluation run is a batch of questions given to the agent as a dataset. Each run shows the question asked, the prompt, and whether the agent resolved or didn't resolve it. Select a run to open the conversations from that dataset and review the agent's responses in detail.

For more on creating and using evaluation runs, see Run evaluations to assess agent performance.

Review test conversations

The Test tab shows conversations you generate yourself using the test sidebar while you're configuring the agent. Use it to check how the agent handles specific questions before customers see it in action.

Filter conversations

Each tab keeps its own filters, so the view you set on one tab doesn't change another. You can filter by:

• Last updated

• Customer feedback

• Review status

• Date range, such as the last seven days

Use filters to focus on the conversations that need your attention - for example, recent conversations a customer marked as unhelpful.