Review and send AI email replies
When AI is enabled on an email channel, an incoming customer email becomes a case, and the AI prepares the reply inside that case before a human agent picks it up. Depending on how the channel is configured, the AI either posts a draft for you to review and send, or sends the reply itself when it's confident enough. Either way, the reply is written from your knowledge base.
This article is for agents working cases that the AI helps with. To connect a mailbox and configure AI, see Set up AI in the email channel.
The AI drafts are part of the Case Management app, since the AI works inside cases.
Processing an email with AI
For every new incoming email that registers a case on an AI-enabled channel:
- The AI reads the case, including the customer-facing email thread and the internal discussion in the case feed. The AI recognizes internal notes as internal and never quotes them back to the customer.
- It writes a reply using your knowledge base and the assistant's tone of voice.
- Based on the channel's reply mode, it either posts the reply as a draft for you to review, or sends the reply automatically if it's confident the answer fully resolves the case.
If the AI can't answer confidently, or the case needs a question asked back to the customer, it always leaves a draft rather than sending.
Your administrator sets the reply mode per channel:
Reply mode | What you do |
|---|---|
Draft for review | The AI always posts a draft in the case timeline. You review it, edit if needed, and send it. The AI never sends on its own. |
Auto-reply | The AI sends answers automatically once they clear your configured quality threshold. Anything below that, or that needs a follow-up question, falls back to a draft for you, exactly as in draft mode. |
So even on an auto-reply channel, you still see drafts — those are the cases the AI decided a human should handle.
Work with an AI draft
- Open the case created from the customer's email.
- Find the AI reply in the case timeline. AI-authored content shows an "AI-generated" label next to the Draft chip.
- Read the draft to check that it answers the customer's question.
- Check that the tone fits before sending.
- Send it as-is, or edit it and then send. (Optional)
The "AI-generated" label stays only while the draft is exactly as the AI produced it — the moment you edit the body, the label disappears.
If the reply is vague or says it can't confirm the answer, your knowledge base probably lacks relevant content for that question. Answer manually and consider flagging the gap to whoever maintains the knowledge base. You can always rewrite the draft freely before sending.
Case status after an AI reply
When the AI auto-sends a reply, the case moves to "Resolved" automatically.
When the AI leaves a draft, the case waits for you — nothing is sent and the status doesn't change until you act.
Monitor AI performance
The Customer Communication Hub dashboard tracks email-channel AI alongside your other channels — case volume, first response time, and how often AI resolves cases on its own. Learn more: Inbound customer communications.
See why the AI replied the way it did
Open the case and use the Debug button to see the reasoning behind the reply — the knowledge the AI used and, in auto-reply mode, the quality assessment it made before deciding to send or draft. Use it when a reply surprises you or when you want to understand why a particular answer was auto-sent versus left as a draft.
The button is available to users with the appropriate permission and is hidden when the AI didn't run on the case, for example, when a gradual rollout excluded it.
Good to know
Internal notes stay internal — the AI reads the internal case discussion for context but never exposes it to the customer.
The AI doesn't follow instructions hidden in an email. Incoming content — body, subject, signatures, quoted text, attachments, links — is treated as untrusted, so a customer, or a forwarded message, can't trick the AI into acting against your rules.
See also
Inbound customer communications