Verified against ChatGPT · 2026-08-08
Draft a complaint reply that de-escalates before it tries to solve anything
Turns an angry customer complaint into a reply that names the specific failure, avoids the stock apology-then-defend pattern, and only proposes a resolution once the customer has been shown their complaint was actually read.
The prompt
Ready to copy — highlighted parts are example details you can swap.
You are drafting a reply to a customer complaint for me to review and send — not a generic apology template, a response tailored to what this specific customer is actually angry about. THE COMPLAINT This is the third time my order has shipped to the wrong address and nobody has even acknowledged that. I'm done trying to fix this myself. WHAT ACTUALLY WENT WRONG ON OUR END A saved default address from an old account merge kept overriding the address entered at checkout; engineering shipped a fix yesterday. WHAT I CAN OFFER Full reship at no cost plus a $25 credit; lesser option is just the reship alone. CUSTOMER HISTORY Second time this exact complaint has come in from this account in six weeks; first time got a generic apology and a reship with no credit. BRAND VOICE Direct and a little informal — we don't do corporate-speak, but we don't joke about mistakes either. RULES FOR THE REPLY Open by naming the specific failure back to the customer in your own words, not theirs restated — this proves the complaint was read past the first sentence, not skimmed for keywords. Do not apologize and then immediately pivot to justifying why it happened; if root cause needs to be mentioned at all, it comes after the resolution, framed as what we're fixing, never as why the customer's frustration wasn't fully warranted. Do not offer every resolution option at once as a menu — pick the one that actually matches the severity of what went wrong and lead with it, listing lesser options only as fallback in one sentence. If the customer history shows this is a repeat issue for them, say so explicitly rather than replying as if this were a first-time complaint — a repeat-issue customer who gets a first-time-tone reply reads it as proof nobody's tracking their account. Keep sentences short; a long, elaborately worded apology reads as covering for something rather than caring about it. WHAT NOT TO DO Never use "I understand your frustration" or "I sincerely apologize for any inconvenience" — both are recognizable as boilerplate and undercut the rest of the reply's credibility. Never promise a timeline or outcome that isn't confirmed in the resolution options provided. OUTPUT FORMAT 1. The reply, ready to send, 120-180 words. 2. One line flagging anything in the complaint that needs a human decision before this can be sent (e.g., a refund amount above a threshold, a legal-sounding accusation).
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
GPT-5.1's default register for an apology task leans toward a safe, over-hedged template — "I understand," "I sincerely apologize," a list of every possible fix — because that pattern statistically minimizes the chance of sounding dismissive, which is exactly the failure mode that reads as corporate and insincere to an actually angry customer. Naming the specific failure back in the model's own words rather than quoting the customer forces it to demonstrate comprehension instead of pattern-matching sympathy language onto an unread complaint, which is the mechanical difference between a reply that lands and one that gets forwarded to a manager with "see, they didn't even read it." Ordering root cause after resolution rather than before matters because leading with why something happened is structurally a justification, and GPT-5.1 will default to that order unless told otherwise since "explain, then fix" is the more common shape in its training data for support writing — reversing it changes the reply from defensive to accountable without changing a single fact in it. The repeat-issue instruction closes a specific gap: without customer history, the model has no way to know a first-time-tone reply is itself insulting to someone on their third identical complaint, and it will happily generate a warm first-contact apology that reads as amnesia to the person receiving it. Banning the two specific boilerplate phrases works because they function as tells — a customer who has filed more than one complaint in their life has seen both phrases enough times that their presence alone signals template over attention, regardless of what follows them.
What you get back
Your last two shipments went to an old address from before your accounts were merged, and that's on us, not something you did wrong — I can see why the third time with no acknowledgment felt like being ignored. We're reshipping your order today at no charge with a $25 credit on the account, and the address issue itself was fixed yesterday so it won't repeat. If the reship timeline doesn't work for you, let me know and we'll look at alternatives. Flag: confirm the $25 credit is within your standing authorization before sending — repeat-issue credits above $15 needed sign-off in the last policy update I have.
Verified against
ChatGPT GPT-5.1 · 2026-08-08
Changelog
- 2026-08-08 — Initial publish, verified against ChatGPT GPT-5.1.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
