Transcripts
Review your AI agent's conversations, filter the ones that need attention, and track review status and notes.
Transcripts is where you review what your AI agent actually said. Every AI conversation is listed with its full message history, and you can filter for the ones that went wrong, mark each one’s review status, and leave notes.
Open it from Transcripts in the left sidebar.
Conversations in Transcripts
Transcripts holds chat conversations and completed voice calls. Chat includes Chatbot (Robylon Chat), WhatsApp, Instagram, Facebook, Twilio Text (SMS), YouTube comments, and chats from a connected Intercom, Freshchat or SalesIQ. Email tickets are not shown in Transcripts; review those in the inbox.
Find conversations that need review
Click the filter icon above the list, set your filters, and click Apply. The most useful filters for finding problems are:
| Filter | Use it to find |
|---|---|
| Transcript Status | Conversations waiting for review: Pending Review or Needs Review. |
| User feedback on messages | Conversations where a customer gave a thumbs-down (Negative) to a message. |
| CSAT Rating | Low-rated conversations. |
| KB answer | Conversations where the agent did, or did not, answer from the Knowledge Base. |
| Chat Net Sentiment Score and Chat Sentiment Shift Score | Conversations where the customer was unhappy, or got less happy. See Sentiment analysis. |
| Evaluations | Conversations an evaluation has run on. |
| Handled By | Conversations handled by the AI alone, or by the AI and a teammate. |
| Test conversations | Your own test chats, or only real ones. |
You can also filter by Date Range, Channel, Human Agents, Agent Groups, Transcript notes, Chat Status, Chat Priority, User Turns, Tags, Intent Classification, Properties & Entities, and, for voice, call direction, status, duration, calling number, voice AI agent and recording.
Track reviewed conversations
Set each conversation’s Status in the Review section of the right-hand panel, and add Notes for anything your team should know.
| Status | Use it for |
|---|---|
| Pending Review | Not looked at yet. Every new transcript starts here. |
| Needs Review | Something may be wrong and someone should look. Conversations rated 3 or lower in a CSAT survey are set to this automatically. |
| In Review | Someone is looking at it now. |
| Reviewed | Done. |
| On Hold | Waiting on something, for example a fix to a source. |
To update many at once, select them with Click to Select and change their status together.
The conversation panel
Next to the messages, the right-hand panel shows:
- Review: the transcript’s status and notes.
- User info: the customer’s details, for example their email.
- Tags and Variables set during the conversation.
- Evaluations: results of any evaluations run on it.
- Credits used: the credits the conversation consumed, the number of turns, and a breakdown by model. See Usage.
Fix a wrong answer from a transcript
For an answer from the Knowledge Base, click Click to Improve under it, correct the question and answer, and save it as a Q&A. See Click to Improve.
For other answers, fix the source:
- An answer from a custom AI agent: edit that agent’s instructions. See Custom AI agents.
- A message from a workflow: edit the step in the workflow that sent it. See Workflows.
Download a transcript
Download a single transcript as a text or PDF file, or select several with Click to Select and download them together.