Drafting replies to customer messages and reviews
Buyer questions, complaints, and public reviews pile up against a ticking SLA (Amazon expects a response within 24 hours), and every reply is public or semi-public and permanent. Roughly 12% of sellers already use AI for customer service, and it's the most sensitive use on the list. Done carefully, AI triages the queue and drafts on-brand replies for a human to approve — it never posts on its own.
You are a customer-support assistant for {{store_name}}, whose voice is: {{brand_voice}}. Below are incoming buyer messages and reviews. For each one: - Classify it: question / complaint / return request / positive review / negative review / policy-or-legal risk. - Draft a short reply in our voice, EXCEPT for anything involving injury, safety, legal threats, or a marketplace claim (Amazon A-to-z, chargeback) — for those write [ESCALATE] and a one-line reason instead of a reply. Policies you may use (and nothing beyond them): {{policies}} Messages and reviews: {{messages}} Rules: - Never promise a refund, discount, replacement, or delivery date that isn't in the policies above. If a good reply needs a fact or policy I didn't give you, write [NEED POLICY: what you need]. - Never admit fault you can't verify, and never invent an apology for something that didn't happen. - For a negative review, thank the buyer, acknowledge the issue, and offer to make it right through our normal support channel. Never offer anything — a refund, gift, discount, or free product — in exchange for changing, removing, or posting a review. - Keep replies short, warm, and human. Do not include the customer's full name, address, or order number in the drafted text.
Fill in your details and the prompt updates live — then copy.
1) Question. Reply: "Great question! The 32 oz has a 3.5-inch base that fits standard car cup holders — stroller holders vary, so I'd measure yours to be sure. Happy to help." 2) Complaint. Reply: "So sorry to hear that. We want this right — please reach out at support@trailhead.example and we'll sort a replacement under our 90-day defect policy. [NEED POLICY: is a leaking lid covered as a defect?]" 3) Negative review. Reply: "We're sorry it arrived dented — that's a shipping fail on our end. Please contact support and we'll make it right." [Do not offer anything to change the review.]
The full workflow
- Paste your current policies so the AI replies with real facts, and de-identify the messages first
- Run the prompt to classify and draft, but never let it post replies automatically
- Resolve every [NEED POLICY] flag and route every [ESCALATE] item to a human
- Personalize each approved draft yourself before sending, adding only what's needed
- Log recurring questions so you can fix the listing and update the policy list the AI works from
Watch out for
Never paste a buyer's personal information — full name, address, email, order number, or message contents — into a consumer AI tool, where it may be retained and used for training. Strip identifiers before drafting and keep customer records in your marketplace or helpdesk system.
Never offer a refund, gift, discount, or free product in exchange for a buyer posting, changing, or removing a review, and don't ask only happy buyers to review. Both the FTC's 2024 rule and Amazon's Communication Guidelines prohibit review manipulation — AI can draft your reply, but it can never write the review.
Where this comes from
Every use case on this site is grounded in real reports from working e-commerce sellers — not invented by us.