Working with tickets

Reply checks: what the "Before you send" warnings mean

Updated

Reply checks: what the "Before you send" warnings mean

On desks with AI triage & assist switched on, Jev reads your reply when you press Send and checks it against the customer's latest message. If something looks off, a Before you send box appears above the Send button. It's a second pair of eyes, not a gate: you can always send anyway.

What happens when you press Send

The button shows Checking… for a moment. Then either:

  • nothing looks wrong, and your reply goes straight out, or
  • the Before you send box lists what Jev noticed. Choose Keep editing to fix it, or Send anyway to send it as it is.

Editing the reply clears the warnings, and it's checked again next time you press Send.

The check never holds you up for long. If the AI is slow or unavailable, your reply is sent without it.

The warnings

"This reply may not answer what the customer asked in their latest message." Jev thinks your reply doesn't address the latest question. Common causes: answering an earlier message, or a holding reply ("we'll look into it") when they asked something specific. Re-read their last message.

"The tone may come across as curt or defensive. Consider softening it." The reply reads as abrupt, dismissive or blaming. Short replies aren't a problem in themselves; this flags ones that could land badly, especially with an upset customer.

"This reply commits to a refund, credit or deadline. Double-check you can honour it." Your reply promises a refund, credit, discount, compensation or a specific date. This one is information, not a problem. It's there so promises are made on purpose.

"The reply mentions an attachment, but nothing is attached." You wrote something like "please find attached" but haven't added a file. Attach it, or change the wording.

"The reply looks like it contains sensitive data (password, card, bank or ID number). Remove it before sending." Your reply seems to include a password, a full card or bank account number, an ID number, or an API key or other secret. Card numbers and South African ID numbers are also caught by a built-in check that works even when the AI is unavailable. Remove the number, or send it a safer way.

Good to know

  • Only public replies are checked. Internal notes aren't.
  • It's advisory. Jev can be wrong, both ways. A reply with no warnings can still be wrong, so keep reading before you send.
  • Your reply isn't changed. The check never edits your text.
  • Turning it off. It comes with AI triage & assist on the desk (Settings → Help Desks → your desk). There's no separate switch for reply checks.

Related: Drafting replies with the AI assistant · Replying to tickets: public replies vs internal notes · Turning on AI triage for a help desk