When AI Goes Wrong and How to Fix It

Let’s be honest. Not every AI result hits the mark.

You’ve probably had ChatGPT write something that sounded a bit… off. Too formal. Too vague. Not what you’d say. Maybe it made things up or misunderstood the job entirely.

That doesn’t mean AI isn’t useful. It just means it needs better direction. And this month, we’re digging into how to spot when AI is wrong, and what to do about it. Because once you know how to guide it, AI becomes more than a tool. It becomes a consistent part of your business.

When Good AI Goes Bad

Here’s where things often go off track:

1. It sounds like a robot

That stiff, overly polite tone. You don’t talk to customers like that, and neither should your emails.

2. It goes off topic

You ask for a follow-up, and it gives you a full sales pitch. Or it includes extras you never mentioned. If you're light on detail, AI starts filling in gaps.

3. It gets the format wrong

You wanted a short update, but you get five paragraphs. It’s not trying to be difficult—it’s just not clear on what you want.

4. It makes stuff up

If you don’t upload the quote, AI will try to guess the details. That’s where hallucinations come in: product specs that don’t exist, finance plans you don’t offer, colours you don’t stock.

5. It always agrees with you

This one’s important. AI is designed to help, not challenge. It will often sound confident, even if it’s wrong. At the same time, it will confidently confirm what you suggest! That’s why you need to give it specifics and sometimes ask it to review its own output.

The Fix: Give It What It Needs

A strong prompt helps. But so does the right supporting info. Don’t just say:

“Write me an email to follow up a quote.”

Instead:

• Upload the actual quote
• Add any notes or email history
• Use your saved GLASS prompt template

This lets the AI write based on facts and not guesses.

Beyond Prompts: Smarter AI Skills

Knowing how to write a good prompt is part of it. But two other skills matter just as much:

1. Ask AI to review its own work

If something feels off, don’t delete it, just ask:

“How could this sound more like a real conversation?” “Does anything here sound vague or unclear?” “What would you remove to make this shorter and sharper?”

You’ll get a refined version, often with solid reasoning behind the changes.

2. Don’t accept the first answer

AI sounds confident, even when it’s winging it. That’s why you should always glance over names, products, numbers and tone. If something feels too generic, it probably is.

Fixing Instead of Starting Over

If the message is nearly there, you don’t need to regenerate from scratch. Try:

• “Rewrite this in one paragraph”
• “Make this more direct and less formal”
• “Remove any fluff and focus on the key offer”

Over time, you’ll build a small bank of these tweaks. Train your team to use them too.

Train Your Team to Spot the Gaps

If you’ve got more than one person using AI (e.g. sales, ops, admin) show them examples of both weak and strong output.

Create a shared folder of:

• Prompts you use regularly (in GLASS format)
• Real before-and-after examples
• Output that needed fixing and how you improved it

This builds confidence and consistency and means you’re not starting from zero every time.

This Isn’t About Perfection, It’s About Progress

No AI tool is flawless. And no one gets it right every time. But if your first reaction to a dodgy output is “AI doesn’t work,” you’re missing the opportunity. AI isn’t the expert. You are. It’s there to help you move faster and get further, with your input guiding it. And the more you use it, the better you’ll get.

We’ve put together a short video inside GlazePro AI that shows how to:

• Upload your own quotes
• Use GLASS prompts effectively
• Review and refine AI output like a pro

You’ll find it here: https://thinkivity.co.uk/GlazePro-AI