Review your AI agent's answers each week
Your AI agent only knows what your published articles and knowledge say. A short weekly look at what it couldn't answer is how its answers get better. Set aside half an hour.
Owners and admins run tests. Pick one owner for this routine, usually whoever runs support.
What to look at
Where | What it tells you |
|---|---|
Reporting > Agent answers | What your AI agent resolved, what it handed off, and the top handoff reasons. |
Reporting > Satisfaction | Survey ratings and comments, worst first. |
Your agent > Conversations | Every conversation your AI agent has had, with the ones it is working on first. |
Each week
Read the numbers
Open Agent answers for the last 7 days. Compare the resolution rate and the handoff rate with the week before, and note the top handoff reasons. A reason that keeps growing is where to start.
Read the conversations behind them
Open the handed-off and low-rated conversations from the report. Each one shows the sources your AI agent used.
Fix the article, not the agent
If an article is wrong, missing or unclear, fix it and publish: drafts are never used. If the topic should always go to a person, add a handoff topic. If it needs steps, write a playbook.
Test the change
In Your agent > Customize, use the Test panel to ask the question again, as an anonymous visitor or as a real customer. A test runs your real AI agent, but stores and sends nothing.
A test always runs as a chat with Normal priority and no labels, so a playbook limited to email, another priority or a label doesn't run in it.
Signs it's working: fewer handoffs for questions your articles should answer, and fewer low ratings.