Tickets AI Drafts: AI-drafted Replies

Review AI-drafted ticket replies in Tickets → AI Drafts: a pending queue with confidence and waiting time, a side-by-side review of the original message and the editable draft, then send, edit and send, reject back to a human, or save the reply to the knowledge base.

When AI triage drafts a reply for a ticket instead of sending it straight away, the draft waits for a person. Tickets → AI Drafts (/tickets/ai-drafts) is that waiting room: its title even says “Pending Review”, with the count of drafts waiting. Nothing on this page goes out to a Client on its own — every draft leaves only when you open it, read it against the original request, and click Send, or you send it back to the human queue with Reject. How a ticket earns an AI draft in the first place — the AUTO / REVIEW / HUMAN triage modes — is covered in the Assignment & AI triage article; this page is about the review itself.

Step 1: Read the pending queue#

• Each row is one waiting draft: the ticket number (TKT-), the ticket subject, and a Review button that opens it.
• Confidence is the AI’s own score for its draft, shown as a colored percentage: green at 80% and above, orange from 50%, red below. Treat it as a hint for how carefully to read, never as permission to skip reading.
• Waiting shows how many whole hours the draft has been sitting unreviewed. At 4 hours and beyond the waiting time turns red and bold — an unanswered draft is a Client still waiting for a reply, so clear the oldest ones first.

Step 2: Review side by side — original on the left, draft on the right#

1. Click Review on a row. The review window opens with the ticket number and subject on top; if this draft has already waited 4 hours or more, a red warning at the top says it is overdue and needs attention.
2. Left card, “Original email”: the customer’s own words, exactly as they arrived. Above them you can see the draft’s confidence and its AI category, plus the “Tools used” — the data the AI looked up while drafting (for example order or tracking lookups), so you can judge whether it answered from facts or from thin air.
3. Right card, “AI draft (editable)”: the suggested subject line and the reply body. Both are plain editable text — fix names, numbers, dates and tone directly in the box; there is no separate edit mode.
4. Check the draft against the original before anything else: does it answer what was actually asked, are order numbers and amounts right, and does it promise only what you can keep? The body cannot be sent empty — if the draft is blank or beyond repair, reject it instead (next step).

This side-by-side review is the whole point of Tickets → AI Drafts (/tickets/ai-drafts): the AI does the looking up and the first writing, and a person who can see both sides at once does the judging. The ticket itself stays in Tickets → List, where the sent reply will appear in its conversation once you approve the draft.

Step 3: Decide — send, edit and send, or reject to a human#

1. Send: if the draft is right as written, click Send. The reply goes out as the ticket’s answer and the draft leaves the pending queue.
2. Edit and send: change the subject or body first, then click Send — what goes out is your edited version, not the original draft. Small fixes (a wrong date, a missing order number, a warmer tone) are normal and expected; you do not need to reject just because you touched the text.
3. Reject → human: when the draft is wrong in a way editing cannot fix — wrong facts, wrong tone for an angry Client, or a question a person should handle — write a rejection reason in the box at the bottom (optional, but a reason helps the AI do better next time) and click “Reject → human”. The draft is discarded and the ticket goes back to being worked by a person.

What happens after your decision#

• A sent draft leaves the pending queue at once and the count in the page title drops — if it is still listed, refresh the page before assuming anything went wrong.
• A rejected draft is gone for good: there is no “un-reject”. The ticket itself is not closed or changed by a rejection — it simply waits for a person to write the reply by hand in Tickets → List.
• Either way, the queue only ever holds drafts nobody has decided on yet — a clean queue at the end of the day means every AI reply that needed a human has had one.

Step 4: Save a good reply to the knowledge base#

Some drafts are worth keeping. In the review window, “Add to KB” opens a small form pre-filled with the ticket’s subject as the title and your current (possibly edited) draft text as the content — adjust both, confirm, and the reply is stored as a knowledge base entry in the draft’s language. Future drafts can then reuse wording that a human already approved. This is separate from sending: adding to the knowledge base does not send the reply, so do both if the Client is still waiting. The entries you save live with the rest of your tenant knowledge — see the Knowledge Base Admin article for how bases are maintained.

Docklify features used in this article#

• AI Drafts pending queue: ticket number, subject, confidence score and waiting time with the 4-hour red flag — Tickets → AI Drafts (/tickets/ai-drafts)
• Side-by-side review: original message, AI category and tools used on the left; editable draft subject and body on the right — Tickets → AI Drafts → Review
• Approve and send, with or without your edits — Tickets → AI Drafts → Review → Send
• Reject back to a human, with an optional rejection reason — Tickets → AI Drafts → Review → Reject → human
• Add to KB: save an approved reply wording to the knowledge base for future drafts — Tickets → AI Drafts → Review → Add to KB