Tone and clarity rewrite for customer replies (L1)
Support and CS paste their own draft reply into the assistant and ask for a clearer, calmer version before sending. The human still writes the answer. AI only fixes structure, tone, and length. This is the lowest risk way to get a support org comfortable with AI.
The steps
- 01
Publish the rewrite prompt
Tool: ChatGPT
One prompt: keep every fact, remove hedging, lead with the answer, cap at 120 words, match our support voice. Include two before and after examples from real tickets with names removed. Owner: support manager. DoD: prompt plus examples live in the support wiki.
- 02
Set the hard rule
Tool: Manual
AI never invents a fact, a date, a refund, or a fix. If the rewrite adds anything the agent did not write, the agent deletes it. This rule is the whole safety model at L1. Pitfall: agents letting the model answer the ticket. That is an L3 play with retrieval, not this one. DoD: rule is in writing and covered in the team meeting.
- 03
Use it on escalations first
Tool: Zendesk
Start with angry or complex tickets where tone costs the most. Those are also the replies managers already review, so you get a natural quality gate. Owner: escalation owners. DoD: every escalation reply for two weeks goes through the rewrite.
- 04
Score readability and reopen rate
Tool: Manual
Compare reopen rate and CSAT on rewritten replies versus the prior month baseline. If nothing moves, keep it as a speed aid and say so honestly rather than claiming a quality win. Owner: support manager. DoD: a one-page result with both numbers and a keep or kill call.
Tools in this playbook
Next playbooks
Unfamiliar terms are defined in the AI and Revenue Dictionary. Related frameworks live in the framework library.