Productivity • Technology

Bunnings’ Chatbot Gave Illegal Advice. Could Yours?

Shocked excited influential mature gray-haired asian, business woman, manager, is sitting at computer, received an unexpected message, looks surprised at computer screen, takes off glasses
Written by Matt Duffin

In December 2025, a customer in Queensland asked the Bunnings AI Assistant how to replace the plug on an extension cord. The handy chatbot gave them six helpful steps.

Bunnings’ chatbot, AKA:Buddy, had just walked a customer through illegal and potentially dangerous electrical work. Bunnings moved quickly, tightening its safeguards so licensed-trade questions now get pointed to a professional, and adding a warning above the chat box.

If it can happen to one of the biggest retailers in the country, it can happen on your website too.

I build AI chatbots for Australian small businesses, so I spend a lot of my week trying to make bots that are genuinely helpful without saying something the business will regret. If you’re thinking about a chat solution of your own, here’s what to check before you put one on your site.

Your chatbot’s words are your words

Plenty of business owners think of a chatbot as a tool, like a search bar. Not really worth putting much thought into it, right?

Customers don’t see it that way. To them, the chat box on your website is your business talking.

The law tends to agree. Under the Australian Consumer Law, it’s the business’s job not to mislead customers.

Air Canada tried it. In 2024, its chatbot told a grieving passenger he could book a full-price flight and claim a bereavement discount afterwards. When he asked for the refund, the airline said no and argued the chatbot was responsible for its own words. The tribunal didn’t buy it and ordered Air Canada to pay up.

The amount was small. But the implications weren’t.

People will try to make it give bad advice

Bunnings’ customer was asking an honest question. Plenty of people aren’t.

One of the bots I built is for an outdoor gear store. A surprising amount of its traffic is people trying to get it to recommend gear that’s completely wrong for their trip. The kind of setup that would leave someone freezing on a mountain.

Some of them are testing it. Some are hoping for a funny screenshot. Either way, the stakes are real. If a bot happily signs off on the wrong sleeping bag and someone actually takes that advice into the alpine, that’s not a bad review. That’s a safety problem with your name on it.

Every business has its version of this. If your products can hurt someone when they’re used wrong, your chatbot needs to know where its advice stops.

People will try to make it say something awful

The same store’s bot also gets people trying to make it say something racist. They’ll feed it an offensive nickname for a place and try to get it to repeat it back.

Why would anyone spend their evening doing that? No idea. But they do, and it’s not rare. Any chatbot on a public website will have someone poking at it for the worst possible answer, and if they get one, it’s your logo in the screenshot.

This is the side of AI chatbots nobody puts in the sales brochure. Building a bot that answers questions is easy. Building one that holds up when people are actively trying to break it is the real work.

The balance is the build

The easy answer is to make the bot overly cautious about everything. But this just results in real customers not getting anything useful from their chat experience.

Also, if you go too far that way the bot hands every second question to a human – which is the opposite of what most business owners want.

Customers notice too. They ask a simple question and get told someone will get back to them.

There’s no universal right setting. The right balance comes from the owner deciding exactly which questions the bot answers, which ones it hands over, and how fast.

And the infrastructure the chat works on needs to be smart enough to enforce it.

5 questions to ask before you put a chatbot on your website

Whether you’re using a free plug-in or paying for custom chatbot development, ask these before it goes live:

  1. Where does it get its answers? It should answer from your own products, prices, and policies, not from whatever it picked up on the internet.
  2. What will it refuse to answer? Licensed work, medical or legal advice, and anything safety-related should have clear limits.
  3. What happens when someone tries to break it? Ask to see it tested with nasty, silly, and dangerous questions, not just the easy ones.
  4. When does it hand over to a human, and how? You should decide this, not the software.
  5. Can you read what it’s saying? You need to see the conversations, so you catch a bad answer before a customer screenshots it.

A chatbot that answers customers at 11pm is one of the most useful things a small business can add to its website. Just remember that every answer it gives, it gives in your name.

About the author

Matt Duffin

Matt is the founder of Matey, which builds custom AI chatbots for Australian online stores, retailers, tradies, and service businesses. He's a former mechanical engineer turned online solopreneur.