The AI That Thanked a Customer for Their "Wonderful Experience" (It Was Food Poisoning)
A mate of mine runs a cafe in Brisbane. Good little spot. Proper coffee, decent brekky rolls, the works.
Last month he rang me up, properly rattled. "Mate, I need you to look at something."
He'd set up one of those AI tools to auto-respond to Google reviews. You know the ones. "Save time! Never miss a review! AI handles it all!" Sounded brilliant. Set it and forget it.
Except the AI forgot one rather important thing: reading comprehension.
A customer left a one-star review. Detailed. Specific. Something about a dodgy egg roll and twelve hours of regret. The kind of review that makes you wince just reading it.
The AI responded within four minutes.
"Thank you so much for your wonderful dining experience! We're thrilled you enjoyed your visit and can't wait to welcome you back soon. Your satisfaction is our top priority!"
His satisfaction was absolutely not the top priority that evening, I can tell you that.
It Gets Worse
My mate didn't notice for three days. Three days of this auto-reply sitting there, publicly visible, underneath a review about food poisoning.
By the time he caught it, two more people had screenshot the exchange and posted it to a local Facebook group. The comments were exactly what you'd expect.
"Their AI cares about you almost as much as their food hygiene."
Ouch.
And here's the thing. He's not an idiot. He's a smart bloke running a decent business. He just trusted the tool to do what it said on the tin.
This Is Happening Everywhere
Before you have a chuckle at my mate's expense, this isn't a one-off. It's happening to businesses all over the place, right now, today.
I talked to a plumber in Melbourne who had the same setup. Customer left a two-star review saying the plumber was 45 minutes late and left a mess. The AI replied with "We appreciate your glowing feedback! Our team takes pride in delivering exceptional service."
Glowing feedback. For a bloke who showed up late and left muddy boot prints across the bathroom.
Then there was a dentist in Perth. Patient left a review mentioning they'd been charged for a procedure that wasn't performed. The AI? "Thank you for choosing us for your dental care! We're so happy you had a positive experience."
Happy? The patient was threatening to call Fair Trading.
Why This Happens
Right, so why does AI get this so spectacularly wrong?
Most of these review response tools work on templates. The AI reads the review, classifies it (positive, negative, neutral), and picks a template. Simple enough in theory.
The problem is most of them are rubbish at classification when reviews are sarcastic, detailed, or anywhere in between.
A review that says "The food was interesting" could go either way. A review that says "Well, that was certainly an experience" could mean anything. A review that's three paragraphs of polite British English before getting to the complaint in the last sentence? The AI reads the first two paragraphs, decides it's positive, and fires off a thank you note.
And most business owners set these tools up once and never check them again. Which is exactly what the marketing says to do. "Set it and forget it!" Great sales copy. Terrible operational advice.
The Maths of Getting It Wrong
Let's do the maths on this, because it's worse than you think.
Your average small business gets maybe 5 to 15 new Google reviews a month. Not a mountain of work. Maybe 20 minutes to respond to all of them personally.
Now, a typical AI review response tool costs $30 to $80 a month. Let's call it $50.
So you're paying $50 a month to save 20 minutes of work.
That's $600 a year. To avoid writing a few sentences every couple of days.
And what's the cost of getting it wrong? One screenshot of your AI thanking someone for their food poisoning experience ends up on local Facebook. That post gets 200 comments. Half of them are people who've never heard of your cafe suddenly associating your name with food poisoning AND tone-deaf customer service.
One new customer a week decides to go somewhere else based on that. At $15 average spend, that's $780 a year in lost revenue. From one bad auto-reply.
You paid $50 a month to lose $780 a year. Proper bargain.
What Actually Works
Look, I'm not saying AI has no role in review management. It absolutely does. But you use it as a tool, not as a replacement for your brain.
Here's what works for the businesses I help set up properly.
Use AI to draft, not to send. Have the AI write a suggested response. Then you read it, tweak it if needed, and hit send yourself. Takes 30 seconds per review instead of two minutes. You still save time. But you also catch the ones where the AI has completely missed the point.
Flag negative reviews for human response. Set up a simple rule: anything under four stars goes straight to your phone as a notification. You respond to those personally. Always. Every time. The AI can handle the five-star "Great service, thanks!" reviews all day long. Those are low-risk.
Never auto-respond to one and two-star reviews. Just don't. The downside is too big and the upside is zero. Nobody is impressed by a fast response to their complaint if the response clearly wasn't written by a human who read their complaint.
Personalise the negative responses. When someone says "the chips were cold," your response should mention the chips. Not "We're sorry your experience didn't meet our high standards." That's corporate speak for "I didn't read what you wrote." Say "Sorry about the chips, mate. That's not good enough and I've had a word with the kitchen. Come back in and the next serve is on us."
Check your auto-replies weekly. If you insist on fully automated responses, at least review them once a week. Five minutes scrolling through your recent Google reviews to make sure the AI hasn't done something embarrassing. That's it. Five minutes to avoid becoming a meme.
The Bigger Lesson
This isn't really about review responses. It's about a pattern I see constantly.
Business owner hears about an AI tool. The marketing says it saves time. They set it up. They assume it works. They never check.
Three months later, the tool has been quietly doing damage. Sending weird emails. Responding to complaints with compliments. Booking appointments at impossible times. Whatever it is.
AI is a brilliant assistant. Genuinely. But it's a terrible manager. You wouldn't hire a new staff member, give them zero training, tell them to handle all customer complaints, and then never check their work for three months.
That's what you're doing when you "set and forget" an AI tool on customer-facing communications.
The Rule I Give Every Client
Before you automate any customer-facing response, ask yourself this question:
"If this goes wrong at 2am on a Saturday and I don't notice until Monday, what's the worst that happens?"
If the answer is "nothing much," automate away.
If the answer is "it could end up on Facebook," keep a human in the loop.
Google reviews, complaint emails, refund requests, social media comments. These are all places where getting it wrong has real consequences. Real, screenshot-able, shareable consequences.
The boring stuff, the admin, the internal notifications, the data entry, the report generation? Automate that to the moon. Nobody screenshots a badly formatted internal spreadsheet.
But customer-facing comms? That's where your reputation lives. And your reputation is worth more than 20 minutes a month.
My Mate's Doing Fine, By the Way
He turned off the auto-responder. Started replying to reviews himself. Takes him about 15 minutes a week.
He replied personally to the food poisoning review. Apologised properly. Offered a full refund and a free meal. The customer updated their review to three stars and mentioned the owner's response was "genuine and appreciated."
No AI could have written that response. Because no AI understood that what the customer actually wanted wasn't a template. They wanted to know a real person gave a toss.
Sometimes the most efficient system is a human who cares enough to spend two minutes writing a proper reply.
---
Want to make sure your AI tools are helping your business instead of quietly embarrassing it? Book a free strategy call and I'll walk you through what to automate, what to leave alone, and how to set it up so nothing ends up on Facebook for the wrong reasons.
---
About Steven Tann: Steven is "The Bloke Who Fixes Your Tech Stack." With over 10 years in the trenches helping more than 7,000 small and medium businesses, he cuts through the guru fluff and builds AI, marketing and automation systems that actually work for real business owners. No jargon. No upsells. Just sorted. Find out more at steventann.com.