Aldi AI Customer Service Response: The Prompt Nobody Deleted

The Aldi AI customer service response that reached a Birmingham shopper on September 14, 2026 carried something no template is meant to show: the prompt itself. Jo Reeves had complained about hairs in a chicken caesar loaded dip. What came back told her how to claim a refund, promised the company took her feedback seriously, and left the line “Make it short and concise but do not overdo it” sitting in the text.

Reeves, from Cotteridge in Birmingham, bought the dip while her partner was in hospital in Worcester for an operation. She found two hairs in it. As she told the BBC, she could just about forgive one, but not two.

The reply she received was competent on paper. It explained the refund route and used the phrase “we’re taking your feedback seriously.” However, nobody had stripped out the drafting instruction, and the whole thing read as machinery rather than a person.

Aldi did not dispute it. A spokesperson said the reply fell well below the company’s normal standards of customer service, and that Aldi had contacted Reeves directly to apologise and resolve the issue with the product.

Aldi has not published a breakdown of where AI sits in its complaints process, and it did not describe the system in its statement. What the stray instruction shows is narrower but still telling: somewhere in the chain, someone asked a model to write the reply, and a human never read it closely enough before it went out.

That pattern is ordinary now. Retailers route high-volume, low-value contacts through generative tools and keep people for escalations. Because of this, the failure here was not the technology choosing bad words. It was the last check that nobody performed.

The detail Reeves kept returning to was not the prompt on its own. It was the missing apology beside it. A refund instruction answers a transaction. An apology answers a person, and the message she got did the first without the second.

She put it bluntly: the company might as well have written “send an email to a customer who is making a mountain out of molehill, give the bare minimum and keep them sweet.” That is her reading, not a quoted instruction from Aldi. Still, it captures why a terse, efficient reply can land worse than no reply at all.

Her conclusion was about trust rather than dip. Aldi presents itself as genuine and authentic, she said, and this was what turned up in her inbox.

Stories like this one rarely turn on a factual error. Instead, they turn on tone, and on the moment a customer realises a machine assembled the warmth in a message. Survey data lines up with that instinct almost every time researchers collect it.

Two independent consumer surveys, which both ran well before the Aldi email surfaced, point the same way.

How shoppers say they feel about AI support

The AnswerConnect figure had climbed from 83% a year earlier, and the hang-up number from 29%. In other words, familiarity is not softening the resistance.

Not wholesale, and the numbers above explain why. When 89% of people want a guaranteed route to a human, removing that route is a commercial decision with a cost attached, not a neutral efficiency gain.

What AI is genuinely replacing is the first layer: order lookups, refund rules, delivery windows, opening hours. Meanwhile, the part that decides whether someone stays a customer, getting a proper hearing after something goes wrong, is exactly the part this story shows machines still fumble.

Almost all of them, at some depth. Banks, airlines, telecoms, food delivery apps and grocery chains run generative replies, triage bots or drafting assistants behind the contact form. Most shoppers never find out, which is rather the point of the design.

The visible cases are the broken ones. A stray instruction, an invented policy, a bot promising a discount nobody approved: these surface because something leaked through the seam. For businesses weighing the same tools, our roundup of AI chatbot platforms sets out what each one actually does with a customer message.

Three things, and none of them is “stop using AI.” First, a human has to read anything that answers a complaint, because the cost of one bad reply is a news cycle. Second, an apology is not an optional field. Third, the prompt is part of the product now, and a prompt that tells a model to keep a customer sweet will eventually reach that customer’s own screen.

Aldi moved quickly once asked, which matters. Even so, the fix arrived after a journalist called, not after the complaint landed. As a result, the reputational damage had already spread by the time the apology did.

If you are choosing tools for your own support desk, start with our guide to the top AI tools for small businesses. And for why automated replies drift over time, read our breakdown of AI performance degradation over time.

Sources

BBC News, SurveyMonkey, AnswerConnect. *Photos: New Aldi store by Philip Halling, CC BY-SA 2.0, cropped and resized; Aldi store, Hexham by Graham Robson, CC BY-SA 2.0, cropped and resized.*