playbook
Draft a reply that sounds like a person
Turn a frustrated customer message into a warm, human reply that owns the mistake, gives one clear next step, and never invents details only you can confirm.
متى تلجأ إلى هذا
A customer is upset and you have to reply in five minutes without sounding like a form letter. The temptation is to reach for corporate filler — "we sincerely apologize for any inconvenience" — which reads as a wall the customer has to climb. This system gets you a reply that owns the mistake plainly, gives one clear next step, and leaves [placeholders] for the private details only you can verify — so you ship a human reply fast without ever letting Claude guess an order number or a refund.
جهّز هذا أولًا
- The customer's message, pasted in or saved as
ticket.txt— strip their full name, email, and account number first; you'll fill those back in yourself. - Your voice in 3 lines: how you say sorry, how formal you are, and 2 phrases you'd never use ("per our policy," "as a one-time courtesy").
- What you can actually offer here — refund, replacement, store credit — and what's off the table, so Claude never promises beyond it.
الـ workflow
-
Read the message back before drafting
Don't ask for a reply yet. Make Claude prove it understood the upset and the ask — that catches a misread before it becomes a tone-deaf reply.
أنت تطلبRead ticket.txt. In 3 bullets, tell me: what actually went wrong for this customer, how upset they sound on a scale of mildly annoyed to furious, and the one thing they most want from us. Don't write a reply yet.ما تحصل عليه A short read-back — "order arrived with a cracked screen; furious, second time this has happened; wants a replacement shipped today, not a refund." If that's wrong, you fix it now, not after a draft sets the wrong tone.
This "read it back to me first" move is the single habit that keeps replies from missing the real complaint.
-
Draft the reply in your voice, with placeholders
Now ask for the reply — but pin the voice and force placeholders so Claude can't fill in private details it has no way of knowing.
أنت تطلبNow draft a reply in my voice: warm and plain, no corporate jargon, no "we apologize for any inconvenience." Own the mistake in the first sentence, give one clear next step, and keep it under 120 words. Use [name], [order number], and [what we're doing] as placeholders. Do not invent any order number, refund amount, ship date, or policy.ما تحصل عليه A short, human reply — "Hi [name], you're right to be frustrated, and I'm sorry — a cracked screen twice is not okay. Here's what I'm doing: [what we're doing]..." — with every private detail left as a bracket for you to fill.
-
Pressure-test the tone and the promise
Have Claude check its own draft against the two ways a reply fails: sounding robotic, or promising something you didn't authorize. Cheaper to catch here than in the customer's inbox.
أنت تطلبRead your draft back critically. Does any line sound like a form letter? Did you promise anything I didn't tell you we offer? Flag both, and give me one alternative for the weakest sentence.ما تحصل عليه A short critique — "line 3 reads canned; the phrase 'rest assured' is filler" — plus a flag if it slipped in a promise, and a warmer rewrite of the flat line.
-
Fill the placeholders and send
This is the human step that can't be skipped. You paste the real details from your own approved system, give it one read aloud, and own the send.
أنت تطلبI've filled in [name], [order number], and the offer is a replacement shipping tomorrow with a prepaid return label. Slot those in and read me the final version exactly as it'll send.ما تحصل عليه The finished reply with your real details in place — ready for one last read-aloud before you, not Claude, press send.
Read it aloud once. If it sounds like something you'd actually say to the person, it's ready.
اجعله ملكك
- **Build the repeatable version:** when the same situation keeps coming up, graduate the best of these into the *Build a canned-response library you'll actually reuse* playbook so you start from a template instead of a blank box.
- **Save the move:** once your drafting prompt is dialed in, store it as a
/replycustom command (see the Playbook's *Features* tab) so a fresh draft is one command, not three prompts. - **Tier the apology:** keep a short note on how you respond to mildly-annoyed vs. furious so the warmth scales with the upset instead of being one flat register.
انتبه إلى
- Never let Claude state an order number, refund amount, ship date, or policy — those are
[placeholders]you fill from your own system. A confident wrong detail to a customer is worse than a slow reply. - Strip the customer's full name, email, card, and account number before pasting — model good [placeholder] hygiene and put the real details back in your approved tool, not the chat.
- Claude drafts the shape and the tone; a human owns the send. The apology has to be yours, and only you can confirm what you're actually offering.
ستحصل في النهاية على A warm, on-voice reply with the mistake owned and one clear next step — placeholders for everything private, and nothing promised that you didn't authorize — in about the time it takes to read the ticket twice.