There is a subtle shift happening in how businesses buy AI. Have you spotted it?
It's the move from the "Ooh, shiny" phase to the "Why is the robot insulting my customers?" phase. I have spent the last few weeks behind the curtain with a few brave souls who decided to let AI run wild in their businesses, and the results are, frankly, magnificent in their stupidity.
We were told AI was going to be the cold, calculating mind that optimised our lives. Instead, it’s turned out to be more like a very eager, slightly drunk intern who has read the entire internet but has never actually been outside.
If you have ever wondered why your chatbot suddenly started offering 90% discounts or why your AI-generated report claims your primary competitor is a sentient head of lettuce, you are not alone. There is a secret world of AI absurdities that nobody mentions in the sales brochures, and it’s time we had a quiet word about what is actually going on.
Why AI Hallucinations Happen in Business
I was sitting with a client last Tuesday who was staring at his screen with a look of pure betrayal. His brand-new, expensive AI lead generator had just sent a series of emails to high-value prospects.
The problem? The AI had decided, for reasons known only to its own silicon soul, that his firm specialized in "Extreme Underwater Knitting" rather than "Bespoke Financial Planning."
This is what the industry calls a "hallucination," which is a very polite way of saying the AI is making things up because it’s bored or confused. It happens because LLMs are essentially just world-class predictive text engines. They don't know facts; they know what word usually follows the previous one.
When you don't give an AI enough context about your business, it fills the gaps with whatever it finds in its training data. If your website is a bit vague, the AI might decide to borrow some personality from a 19th-century poet or a disgruntled Reddit user.
The trick—and this is what the gurus won't tell you—is that AI doesn't need more "intelligence." It needs more constraints. It's like a Labrador; if you don't give it a ball to world with, it's going to eat your sofa. You can find more articles on AI where I discuss how to stop your tech from eating the metaphorical furniture.
Common Examples of AI Failing at Simple Tasks
It is genuinely impressive how a machine that can solve complex physics equations can fail so miserably at being "relatable" to a human being. I have seen AI tools do things that would get a human fired before their first coffee break.
Here are a few of my favourite "innovations" I’ve spotted lately:
- The Over-Eager Closer: A chatbot that was told to "be persistent" and ended up harassing a lead until the lead threatened legal action. The AI’s response? "I understand your frustration, but have you considered our premium subscription for legal professionals?"
- The Creative Historian: An AI writing a company "About Us" page that claimed the CEO was a founding member of the Knights Templar. He’s 42 and from Reading.
- The Mirror Whisperer: Someone used AI to "scrub" their LinkedIn profile to look more professional. It replaced his profile picture of him at a BBQ with a generated headshot where he had six fingers on one hand and a subtle third ear.
We are currently in the "uncanny valley" of business automation. We want the efficiency of a machine but the soul of a human, and we’re currently getting a machine that’s trying to wear a human’s face as a mask. It’s a bit creepy, a bit funny, and entirely avoidable if you know which buttons to stop pressing.
How to Fix AI Performance Issues with Better Context
If you want to stop the madness, you have to treat AI like a new employee who has zero common sense but a very high reading speed. Most people fail because they give the AI a one-sentence prompt and expect a masterpiece.
Think of it like this: if you ask a chef to "make food," you might get a sandwich or you might get a slow-roasted shoe. If you want the lobster thermidor, you have to be specific about the ingredients, the temperature, and the fact that shoes are off the menu.
To get better results, you need to provide what I call a "Knowledge Base Railing." This involves:
- Uploading actual company PDFs and case studies as the primary source.
- Explicitly telling the AI what it is not allowed to talk about.
- Giving it a "persona" that isn't just "helpful assistant" (which is code for "boring robot").
The best AI implementations I’ve seen are the ones where the owners were slightly paranoid. They checked the outputs, tweaked the "temperature" settings, and didn't just set it and forget it. If you're feeling overwhelmed by the technical jargon, you might want to book a consultation to see how we can straighten out your digital intern.
The Absurdity of AI Productivity Hacks
I saw a post the other day where a "expert" suggested using AI to attend meetings for you. The idea is that the AI sits in the Zoom call, takes notes, and you go play golf or whatever it is people do when they aren't working.
I tried this. Not the golf, just the AI meeting attender.
The result was two AIs talking to each other for forty-five minutes because the other person on the call had also sent their AI. They reached an agreement to "sync up next week," and neither of us humans had any idea what was actually discussed. We are literally building a world where robots have meetings with other robots to discuss things that will never happen.
It’s the ultimate productivity hack: total, automated stagnation.
The irony is that while we're using AI to save time, we're spending all that saved time fixing the weird things the AI did while we weren't looking. It's a bit like buying a robot vacuum that occasionally decides to repaint your hallway in shades of "mud."
Practical Steps for Successful AI Implementation
If you want to actually benefit from this stuff without becoming a cautionary tale, you need a plan that involves more than just a ChatGPT subscription and a dream.
Successful AI implementation looks less like a "launch" and more like a slow, slightly suspicious integration. Here is how the pros (and people who don't want to look silly) do it:
- Start Small: Don't let an AI talk to your customers until it has spent a week talking to you and your team. If it annoys you, it will infuriate them.
- Verify Everything: If the AI gives you a fact, assume it’s a lie until you check Google. AI is a confident liar.
- Keep the Human in the Loop: AI should be a co-pilot, not the captain. It can write the first draft, but a human needs to add the "not being a weirdo" filter.
- Use Specific Tools: Generic models are okay, but tools designed for specific tasks (like sales or coding) usually have better guardrails. Since it's Thursday, you can get the free book which dives into how to actually use these tools without losing your mind.
The goal isn't to replace the human element; it's to automate the boring bits so the humans can do the things robots aren't good at—like having a sense of humour or knowing when a joke has gone too far.
Conclusion: Embracing the Chaos
We are living through a weird, wonderful, and occasionally terrifying experiment. AI is the most powerful tool we’ve ever had, but it’s currently in its "toddler with a permanent marker" phase. It has the potential to do great things, but it’s probably going to draw on the walls first.
The winners won't be the people who use AI for everything. They will be the people who know exactly when to turn it off.
Next time your AI does something completely mental, don't get angry. Take a screenshot, have a laugh, and then fix the prompt. After all, if the robots were perfect, we'd all be out of a job, and I’d have nothing to write about on a Thursday afternoon.
We're all just figuring this out as we go. Just make sure your "About Us" page doesn't mention the Knights Templar, and you'll be ahead of at least half the competition.
---
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.