When the AI finds that an agent's answer contradicts your knowledge base, it files a flag and leaves a private note in that conversation. Reviewing those flags is a short, regular job: decide what counted, fix what was wrong, and move on.
Where to find them
Go to Team → Agent monitoring. The menu entry carries a number — the flags nobody has looked at yet. The Flags tab opens on four figures for the period you are viewing:
- Turns checked — how many agent replies the AI reviewed
- Unhandled — waiting for you; this is the number in the menu
- Confirmed inaccuracies — reviewed, and the agent really was wrong
- Closed — reviewed, and decided not to count
Unhandled flags always sort to the top. You can narrow the list by date, agent, brand and status.
Deciding on a flag
Each flag shows the agent's words beside what your knowledge base says, plus the source it came from. Read the source before deciding — the View passage button shows the exact text the AI judged against.
- Confirm inaccuracy — the agent was wrong. The flag still counts in reports, and the private note stays in the conversation marked as reviewed.
- Close — not an inaccuracy — it should not count. The private note is removed from the conversation and the flag stops counting.
You can add a short note explaining your decision, and nothing is permanent: Reopen puts a flag back on the pile, restoring the note if closing had removed it.
When the knowledge base was wrong
A flag often means your documentation is out of date rather than that the agent misspoke. The passage view marks this for you:
- Current copy — your knowledge still says exactly this.
- Changed since — the source has been edited or removed since the flag was raised.
Use Open source to edit to jump straight to the article, documentation page or product entry and fix the wording. Fix first, then close the flag — otherwise the same answer gets flagged again tomorrow.
When the agent disagrees
Agents can dispute a flag directly from the note in their conversation, and a flag they have replied to is badged Disputed by agent. Open Details & comments to read the thread and reply. Treat a dispute as useful signal — it usually points at documentation that needs updating.
Per-agent view
Opening an agent from Team → Reports → Live chat shows an Agent monitoring (last 30 days) card: how many inaccuracies they have, how many are still unreviewed, and across how many checked replies. "Inaccuracies" counts unhandled plus confirmed flags — anything you closed is excluded, so false alarms never count against anyone.
Getting set up
Agent monitoring is included in the Growth plan and available as a paid add-on on every other plan, the free one included — see Plans and pricing. Once it is on, give the Agent monitoring permission to whoever should review flags (see the permission reference), then choose who is monitored on the Rules tab.
Agents on the receiving end should read When the AI flags your answer.