AI Analysis of Conversations

Alongside every AI-generated reply, WebSpeaker shows a compact analysis of how the assistant understood the visitor’s message. This gives you insight into what the AI thought the visitor wanted, whether it decided a human should step in, and other signals that are useful when you are reviewing conversations. This page describes the analysis addons; for the conversation list, message history, and feedback, see Review Conversations.

Where to Find It

Open a conversation in the Conversations section, then look below any AI Assistant message. When analysis is available for that reply, WebSpeaker shows a row of small labelled buttons — for example Analysis, Handoff, Support, Sources, and Language — next to the message. Click a button to expand its details; click it again to collapse it. The Sources panel is the same source-citation list described in Review Conversations.

Quick Summary Badges

Before you open any panel, a row of short badges above the buttons summarizes the most useful facts at a glance. Depending on the conversation, you may see the detected intent, the journey phase (for example whether the visitor is still deciding or is already a customer), a note that the visitor asked to be connected with a human, or that frustration or a repeated problem was detected.

Analysis Panel

The Analysis panel explains how the assistant read the visitor’s message. It can include the detected intent, the answer mode used to reply (for example an answer grounded in retrieved content, a direct reply, or a handoff to your team), the current topic and the visitor’s goal as understood by the assistant, and a short reasoning note explaining that choice. It may also show the standalone, context-resolved version of a follow-up question, which is useful for checking that the assistant correctly understood a question that only makes sense together with earlier messages in the conversation.

Handoff Panel

The Handoff panel tells you whether the assistant decided the visitor should be connected with a human, and if so, what information is still missing before that can happen (for example contact details or a description of the topic). Once a handoff has actually been submitted to your team, this panel also shows the recorded handoff details: when it was submitted, the topic and summary, and the contact information the visitor provided. Reviewing this panel helps you check that handoffs are triggered at the right moments and that the details passed to your team are accurate.

Support Signals

For conversations that touch on a problem or support request, a Support panel can surface additional signals, such as whether the visitor appears to be reporting the same issue again, how many times they say they already tried to resolve it, and whether the assistant considers the conversation ready for escalation to a human. These signals are particularly useful when reviewing Live Support conversations that started as an AI conversation before being handed over.

Using the Analysis When Reviewing Conversations

Use the analysis together with the message history and source citations to understand not just what the AI answered, but why. If the detected intent or topic does not match what the visitor actually asked, that is a sign the assistant misread the question. If handoffs are triggered too often, too rarely, or with missing details, use the Handoff panel to see exactly what was detected and adjust your content or system prompt accordingly. Combined with visitor feedback, the analysis panels give you a faster way to spot conversations worth a closer look.