Which tool to choose
Only one thing matters: writing in a human tone under pressure.
- Claude tends to convey empathy well and to avoid the canned-responder phrases: useful when the customer is already worked up.
- ChatGPT and Gemini work perfectly fine and, in the versions with web search, also help you verify a policy or a rule the customer cites (for example the return window required by law).
For complaints there's a hidden benefit in using AI: it writes the first version while you're still on edge, and that filter between your gut and the keyboard often saves the relationship with the customer.
How to do it
From computer or app the path doesn't change: you paste the complaint, give the rules, get the draft.
- Gather the context in one line. What actually happened, whose fault it is, what you can offer to make it right. If you don't say it, the AI makes it up.
- Ask for the reply with the right constraints. The working syntax:
Act as a customer service manager, patient and professional. Below is a customer's complaint. Write me a reply in English that: acknowledges the specific problem (not generic apologies), shows you've understood the point, proposes one of the solutions I give you and closes with a clear next step. Warm but understated tone, short sentences, no canned formulas. Solutions I can offer: partial refund, product replacement, discount on the next order. The complaint: "..."
- Reread and correct the facts. AI doesn't know what really happened: check dates, amounts and promises before sending the email.
- If you need to lower the tension, ask for a second version "shorter and warmer." Then you send it.
A concrete example
Luca has a small online sporting-goods store. A customer writes furious: the racket arrived with the strings loose, "disgraceful service, I want my money back." Luca is tempted to reply in kind. Instead he pastes the message into Claude with the prompt above and lists the replacement or the refund as solutions.
AI produces a reply that opens by acknowledging the precise defect (the loose strings, not a generic "we're sorry for the inconvenience"), apologizes once, offers the replacement with free shipping and, as an alternative, the full refund. Luca adds the order number and a line of his own, and sends it. The customer accepts the replacement and leaves a positive review about the recovery. The calm draft defused the confrontation.
When it does NOT work (and how to fix it)
If the customer is right and the mistake is yours
AI can package a perfect reply that, however, dances around your responsibility. When the mistake is clearly yours, write it in the prompt ("the mistake is ours, admit it openly"): a real apology is worth more than ten diplomatic phrases, and the customer recognizes it.
If the reply sounds fake
Sometimes a text comes out too polished, call-center style. Ask the AI to rewrite it "the way a real person would say it, in plain words" and remove a couple of adjectives yourself. Perceived sincerity counts more than perfection.
If the complaint is public (review, social media)
Here everyone reads the reply. Ask the AI for a short response that doesn't argue and that takes the conversation private ("write to us in a direct message so we can sort it out right away"). Never pin the blame on the customer in front of others: even when they're wrong, the public roots for whoever stays calm.
A tip from someone who really uses it
Build a small library of three or four of your own already-approved replies (for delays, defective products, misunderstandings) and use them as an example in the prompt: "reply in this style." AI imitates your way of speaking instead of inventing one, and the replies become recognizable as yours, not churned out by a machine.
Frequently asked questions
Can I have AI reply to complaints automatically?
Technically yes, with a chatbot, but on complaints it's risky: an already-angry customer who realizes they're talking to a robot gets even more furious. Better to use AI as a quick draft that you reread and send, at least for the sensitive cases.
Can AI invent promises I can't keep?
Yes, and it's a concrete danger: if you don't tell it what you can offer, it can promise refunds or timeframes you can't sustain. That's why in the prompt you list the possible solutions yourself, and the AI only chooses among those.
Isn't using AI for complaints dishonest toward the customer?
No, and here's the misunderstanding to clear up: AI doesn't fake an empathy you don't have, it helps you express the one you do have when anger or haste get in the way. The judgment, the choice of solution and the responsibility stay yours. The tool writes the sentence; the will to make it right you supply.