The Dangers of Trusting AI Blindly: Two Stories From My Own Business

Joe Sack · Blog

When did you last do something because AI told you to, only to realise hours later it was wrong? If that question makes you uncomfortable, it should. I run an AI consultancy. I have been working with these tools every day for years. And in the last few months alone, I have wasted hours following AI instructions that were confident, detailed, and completely wrong.

Quick answer: AI tools give confident answers even when they are wrong, because they don’t know what you already have or what “simple” looks like for your situation. Two AI models gave me the same wrong fix for a website problem, and a separate AI wasted two hours of my time on a task a plugin already installed on my site could do in seconds. The fix isn’t avoiding AI. It’s questioning it before you act on it.

Not wrong in an obvious way. Wrong in the way that only becomes clear after the work is already done. This article is about what actually happens when you follow AI blindly, why the “I hired Claude as my SEO Executive” posts on LinkedIn leave out the part that matters, and the one habit that separates people who get real value from AI from people who just get burned by it.

Real Example

What Happens When Two Different AI Tools Give You the Same Wrong Answer?

I was building my new website and hit a URL redirect issue. Not an unusual problem, and not one that should have taken long to fix.

I asked Claude. The answer: duplicate the entire website and point one version to the other. That seemed extreme, so I asked Gemini. Same answer. Duplicate the website.

Two different tools, two different companies, the same instruction. If I’d been less experienced, two independent sources agreeing would have felt like proof it was correct. It isn’t. I pushed back and told both that duplicating a whole website seemed disproportionate for a redirect problem. Only then did they land on the real answer: a few lines of configuration, five minutes of work.

Neither AI stopped to ask whether there was a simpler way. They just gave me steps that would eventually work, without checking if it was the right approach.

I had to talk two AI tools out of the wrong answer. Would you have known to push back?

Real Example

Why Did Fixing a Table of Contents Take Two Hours It Didn’t Need To?

A few weeks later, building a blog post, the table of contents links weren’t jumping to the right sections. I asked for help.

The instruction: go into each heading, find the HTML Anchor field, and type in an anchor ID by hand. Fifteen headings, one by one. I did it. Then came a new problem: duplicate IDs on the list items, which meant going into the raw HTML to strip them out manually. I was about to start.

Something made me stop and ask: is there a simpler way I’m missing? Yes. Delete all of it and use the Rank Math Table of Contents block instead. It handles anchors automatically, in about ten seconds.

Rank Math had been installed on my site the whole time. The AI had no way of knowing what tools I already had, and I didn’t think to mention it. Two hours lost because neither of us asked the right question first.

Why This Happens

Why Does AI Sound Right Even When It Isn’t?

There’s a word for this in AI circles: hallucination. It sounds dramatic. In practice it’s mundane and dangerous. It’s when AI states something false as if it were true. No flag, no hesitation, no “I think” or “let me check.” The wrong answer arrives in exactly the same tone as the right one.

With a human expert, you pick up on tells. A pause. A hedge. A shift in tone when they’re less sure. AI doesn’t have those. And here’s the part most posts about AI leave out: two AI tools agreeing doesn’t make something true. Claude and Gemini gave me the same wrong answer. AI systems can share the same training biases and the same tendency to reach for a complex fix when a simple one exists. Agreement isn’t validation.

If you’re using AI to make decisions in your business — pricing, strategy, technical, legal, financial — this is the most important thing to understand before you start. It’s a theme I come back to often when I talk about how UK small businesses are actually using AI right now.

A Word of Caution

Should You Trust the “I Hired Claude as My SEO Executive” LinkedIn Posts?

You’ve probably seen them. “I replaced my copywriter with AI.” “Here’s how I built an AI team of six for £0 a month.” They do well. Thousands of likes. They feel like insider knowledge, and the people writing them have usually had a genuinely useful experience.

But here’s what they leave out. The person writing them usually has enough background to catch the mistakes. They know when the SEO advice is wrong because they’ve done SEO. They know when the copy needs fixing because they’ve written copy. The AI gave them a fast first draft, not an independent decision. If you’d like a grounded look at what these tools are actually good for, my page on AI explained simply is a good starting point.

A small business owner with no SEO background reads that same post and thinks: great, I’ll just hand this over. That’s a website optimised for the wrong terms for six months. That’s customer emails going out with errors in. That’s a decision made on numbers AI got wrong. The posts show the highlight reel. They don’t show the checking, or the years of experience quietly catching what the AI gets wrong.

Data & Privacy

Is It Safe to Put Your Business Data Into AI?

The posts that worry me most are the ones involving business data. “I uploaded our client list and asked AI to segment it.” “I pasted the email thread with my client and asked AI to draft a response.”

The answer isn’t “never put data into AI.” That would be too blunt, and it would stop you doing genuinely useful things. The real question is which tool, with what data, under what safeguards. The free version of ChatGPT has different privacy terms to a paid business tier. Pasting client data into it is a GDPR question as much as a technology one. A healthcare tool built with proper encryption for clinical notes is a different matter entirely to a general-purpose chatbot.

The LinkedIn posts don’t make this distinction. If you’re unsure whether a tool is safe to use with your data, that’s exactly what a one-to-one AI consultation is for.

The Fix

How Do You Use AI Without Getting Burned By It?

✓	Always ask if there’s a simpler way. Before following a set of instructions, ask what tools you already have that could do this. One question would have saved both stories above.
✓	Push back when something feels disproportionate. I had to talk two AI tools out of duplicating my website. AI responds well to being challenged, it just won’t volunteer the simpler answer unless asked.
✓	Agreement is not validation. Two AI tools giving the same answer doesn’t mean it’s correct. Apply your own judgement, especially on anything consequential.
✓	Know what you’re putting into it. Check a tool’s data policy before pasting anything sensitive in. Free and paid tiers are not the same thing.
✓	Read confident LinkedIn posts with scepticism. They usually show someone who already knew enough to catch the mistakes. That doesn’t automatically transfer to your situation.

The businesses getting the most from AI right now aren’t the ones moving fastest. They’re the ones who slowed down long enough to understand what they were using before they used it. If you want a practical, no-jargon introduction to that way of working, my free AI workshops for North London businesses cover exactly this.

I’ve had to learn this the hard way, twice, in the same month. I’m telling you so you don’t have to.

FAQ

Common Questions About Trusting AI in Business

Joe Sack is an AI consultant based in Crouch End, North London. He helps small businesses and independent professionals use AI without the overwhelm.