I've been sitting on this idea for weeks. Time to share it.
We are currently living through the Great AI Pretend. It is that specific moment in history where every CEO is nodding sagely about "neural architecture" while their actual experience of AI is a chatbot telling them that a pebble is a delicious snack if you sauté it in butter.
Between us, the industry is currently held together by digital duct tape and a lot of prayer. We’ve been told these tools are the second coming of the steam engine, but right now, they feel more like a very confident intern who has been drinking double espressos and reading Wikipedia at 3 AM.
What are AI hallucinations in business?
If you've ever asked an AI for a case study only for it to invent a Fortune 500 company that doesn't exist, you've met a hallucination. It’s the ultimate "fake it 'til you make it" move, except the AI never actually makes it. It just keeps faking with increasing aggression.
The technical crowd will tell you it's a "probabilistic outcome of language modelling," which is a fancy way of saying the machine is guessing what word comes next based on math. Sometimes that math decides the next logical step in your legal brief is to cite a law passed by King Henry VIII regarding the sale of goats.
I recently watched a friend use a popular tool to summarise a board meeting. The AI decided, for reasons known only to its silicon soul, that the CFO had resigned to become a professional kite flier. He hadn't. He’d just mentioned it was windy outside.
Why do AI tools make things up?
The problem is that AI is designed to please us. It hates saying "I don't know." It would much rather lie to your face than admit it’s stumped. It’s like a waiter who doesn't want to tell you they’re out of the sea bass, so they just bring you a plate of wet socks and hope you're too distracted to notice.
When you ask an AI for "innovative ways to use AI for sales," it doesn't search for truth. It searches for patterns. If the pattern suggests that people like lists of ten things, it will give you ten things. If it only knows seven real ones, it will simply hallucinate three more.
I once saw a proposal where the AI suggested the company "leverage quantum collaboration to optimise the coffee machine." It sounded impressive until you realised it was complete bollocks. If you want to see how we actually handle these things properly, you might want to book a consultation before your AI starts firing your best clients.
Common examples of AI fails in the workplace
I’ve been collecting these like rare stamps. There is a certain joy in seeing a multi-billion dollar algorithm go completely off the rails.
- The Invented Expert: An AI once provided a glowing testimonial from a "Dr. Aris Throttlestein." A quick search revealed the doctor didn't exist, but the AI had even generated a fake university for him to work at.
- The Math Problem: Asking an AI to do basic accounting is a thrill ride. I saw one insist that 15% of £100 was £22.50 because "tax was included in the logic." It wasn't. It just liked the number 22.
- The Ghost Product: A marketing agency used AI to write product descriptions for a hardware shop. The AI invented a "self-levelling hammer." People actually tried to order it.
The danger isn't that the AI is stupid. The danger is that it's incredibly articulate while being wrong. It doesn't stutter when it lies. It uses perfect grammar to tell you that the moon is made of brie. For more grounded takes on technology, check out more articles on AI.
How to prevent AI hallucinations in your workflow
You can't fully stop a machine from dreaming, but you can certainly wake it up. If you're going to use these tools, you need to treat them like a posh teenager who thinks they know everything because they’ve been to a TED talk once.
- Lower the Temperature: Most AI models have a "temperature" setting. High temperature means the AI gets "creative" (starts lying). Low temperature means it sticks to the script.
- Provide the Truth: Don't ask it what's in a document. Give it the document and say "Only use this." If you leave the door open, it will wander off into the woods.
- The "Are You Sure?" Paradox: Sometimes, simply asking the AI "Is this actually true?" causes it to have a crisis of conscience. It will often reply, "I apologise, I made that up. Here is the real answer."
- Fact-Check Everything: If an AI gives you a statistic, a quote, or a legal citation, assume it's a lie until proven otherwise.
It's particularly important when dealing with outreach. If you're using SalesM8 to handle your lead generation, you'll find that having a structured system prevents the "creative" outbursts that happen when you just let an LLM run wild in your inbox.
The absurdity of "AI-driven" everything
We have reached peak buzzword. I saw a "smart toaster" recently that claimed to use AI to "understand bread." It’s a toaster. It has two settings: "Bread" and "Fire." It doesn't need a neural network to figure out a crumpet.
The same thing is happening in business. We are over-complicating things that were perfectly fine before. We’re replaced simple, effective human communication with "optimised outputs" that sound like they were written by a Victorian ghost who had a stroke.
The most successful people I know aren't using AI to replace their brains; they're using it to clear the boring stuff out of the way so they can actually use their brains. They don't let the AI drive the car; they just let it wash the windows.
Finding the balance between tech and truth
The real secret—the one the "gurus" won't tell you while they're trying to sell you a £2,000 prompt engineering course—is that AI is a tool of convenience, not a tool of accuracy.
It's brilliant at turning a bulleted list of messy thoughts into a polite email. It's terrible at being a source of absolute truth. If you treat it like a very fast, slightly drunk researcher, you'll be fine. If you treat it like an oracle, you'll eventually find yourself explaining to a client why your "AI-generated strategy" involves a partnership with a company that went bust in 1994.
Let's keep the human in the loop, shall we? At least humans have the decency to look embarrassed when they're caught making things up. AI just stares at you with its blinking cursor, waiting for you to say "thank you" for the nonsense it just served up.
Stay sharp, keep your eyes on the data, and maybe don't buy that self-levelling hammer just yet.
---
Want more stories like this? I share observations about AI, business, and life on the narrowboat at steventann.com.
---
About the Author
Steven Tann is an AI consultant, author of "You're Selling AI Wrong", and founder of SalesM8. He writes about AI, sales, and running a business from a narrowboat on the English canals. Connect with him at steventann.com.