AI Made This Worse
The complaint reply that sounded fake
Written by AI Tools AcademyLast checked on 27 September 2026
Helpful first: Amber: personal or commercial, but redactable
This is a fictional example. Fernway Group is a made-up UK company, and everyone in it, including the customer, is invented.
Complaint replies are one of the most common things people ask AI to help with, and for good reason: when someone is angry with you, a calm first draft is useful. But an angry customer reads every word looking for signs that nobody is really listening, and AI drafts are full of exactly those signs.
Before
Sam Price, Account Manager for the South, received this from the office manager at Cobbold & Wren Surveyors, a long-standing customer:
This is the third time in two months that your engineer has not turned up
in the slot we were given. Each time I have rearranged the office around it.
Nobody called to say he wasn't coming. I've been a customer for six years
and I'm now looking at other suppliers. I'd like to know what you are
actually going to do about it.
Sam checked the job log. Two of the missed visits were caused by the engineer being sent to an overrunning job, and nobody updated the customer. The third was a booking error at Fernway's end. Sam spoke to Tom Elliott, the Sales Lead, who agreed Sam could offer the next quarterly service free of charge and a direct phone call from the engineering coordinator the day before every visit from now on.
Sam felt defensive and didn't trust the first draft to come out right, so asked the AI tool for help.
The AI approach
Sam replaced the customer's name and the firm with placeholders, then pasted the complaint:
Write a professional, empathetic reply to this customer complaint. We want to keep them as a customer. [the complaint, with the name and firm replaced]
Why this works: This is the version that caused the problem: it gives the tool the customer's anger but none of the facts or the agreed remedy, so it has to invent both.
What went wrong
Dear [Customer],
Thank you for taking the time to share your feedback, and please accept my sincere apologies for the frustration and inconvenience you have experienced. I completely understand how disappointing it must be when our service falls short of your expectations, especially as such a valued long-standing customer.
Please rest assured that we take your concerns extremely seriously. I have escalated this matter internally and we are conducting a thorough review of our scheduling processes to ensure this never happens again.
As a gesture of goodwill, I would like to offer you a full refund of your most recent service charge, along with a dedicated account manager who will oversee all future visits personally.
We truly value your business and look forward to continuing to serve you.
Warm regards,
Sam
Sam spotted the refund line and cut the goodwill paragraph, then sent the rest with light edits. The customer's reply came back within the hour: "Did a person write this?"
Several things went wrong at once:
- It never said what happened. The customer asked a direct question, "what are you going to do?", and had said nobody called. The reply doesn't admit the missed calls, doesn't mention the booking error, and doesn't say what will be different next time.
- It is built from stock phrases. "Thank you for taking the time to share your feedback", "rest assured", "we take your concerns extremely seriously", "valued customer". Each is harmless alone. Together they read like a script, and a customer who has already been let down three times hears a script as another brush-off.
- It invented a remedy. A full refund and a dedicated account manager were never agreed. Sam already is the account manager. Tom had approved a free service and a call before visits, which the draft doesn't mention. Sam caught this one. Had it gone out, Fernway would either have had to honour an offer nobody approved or go back on its word in writing. And with the invented remedy removed, the reply Sam sent offered nothing at all.
- It promised the impossible. "To ensure this never happens again" is a guarantee no one can give about engineers and traffic.
The AI was asked for empathy and a retained customer, with none of the facts. So it produced the general shape of an apology and filled the gaps with the most common remedies it has seen. That is what a tool does when the facts it needs are missing.
A better approach
The fix is to give the tool the substance and keep the AI's job narrow. Sam should decide what to say, and use the AI to help say it clearly and calmly.
Sam's own notes, written first, might be four lines: we missed three visits; twice the engineer was diverted and nobody rang you, once we booked it wrong; from now on the coordinator will call the day before every visit; your next quarterly service is free. That is the reply. The AI's job is to turn it into sentences.
Weak
Write a professional, empathetic reply to this customer complaint. We want to keep them as a customer.
Better
Draft a reply to the complaint below using only the facts and remedy in my notes. Say plainly what went wrong, in the first paragraph. Don't add any offer, promise or apology I haven't written, and avoid stock phrases such as rest assured, valued customer or we take this seriously. Keep it under 150 words, direct and human, as if written by the person who looks after their account. My notes: [what happened] [what we will change] [what we are offering]
Why it works: The tool gets the facts and the approved remedy, is told what it may not invent, and is steered away from the phrases that make a reply sound scripted.
Then read the draft as the customer would. A useful test: could this reply have been sent to any customer about any complaint? If yes, it needs more of the specifics that only you know. The strongest complaint replies usually contain one sentence that is plainly true and slightly uncomfortable to write, such as "Nobody rang you, and that was our failure."
On data: customer complaints often contain names, contact details and account history. Replacing them with placeholders, as Sam did, is the right instinct. Your organisation's policy and approved tools still take priority.