Trust, but verify

Every habit so far in this course has been about getting a better answer out of AI. This one is different, and it matters more than all the others put together: never let an AI-drafted fact go out the door, to a customer, a supplier or a regulator, without checking it yourself first.

Confident is not the same as correct

AI writes every sentence in the same fluent, sure-of-itself tone, whether it’s dead right or completely wrong. There’s no hedging, no “I’m not certain”, no visible difference between a fact it knows solidly and one it’s essentially guessed at. That’s what makes it dangerous in exactly the moments it feels most trustworthy.

This isn’t a rare glitch. It’s how the tool works. It’s built to produce the most plausible-sounding next sentence, and a plausible-sounding wrong number reads exactly the same as a plausible-sounding right one.

Where this bites hardest for a trades or SME business

Not every AI-drafted mistake matters the same amount. A slightly off adjective in a social caption costs nothing. These four categories are where it genuinely costs money or trust:

• Prices and figures. Ask it to help draft a quote and it may confidently invent a number, especially if you didn’t give it your actual pricing. Always check the figure against your own price list before it goes to a customer.
• Dates and availability. “We can start Monday” is only true if you actually can. AI has no idea what’s in your diary.
• Regulations and technical specs. Building regulations, electrical standards, planning rules, anything with a legal or safety consequence. These change, vary by region, and are exactly where a wrong-but-confident answer does the most damage.
• Anything about a specific customer. Names, order history, what was already agreed. If you didn’t give it that information this chat, it doesn’t actually have it, however sure it sounds.

For more of the everyday habits that trip people up, our guide on common prompting mistakes has a fuller before-and-after for each one.

The habit itself

Read every AI draft the way you’d read a quote a new apprentice had drawn up: probably fine, worth your own five minutes before it leaves the building. For anything with a number, a date or a regulation in it, check that specific detail against your own records or the actual current rule, not against how confident the sentence sounds.

This one habit is the difference between AI saving you time and AI creating a problem you now have to apologise for. It costs a few minutes. Sending a wrong price or a wrong regulation to a customer costs a great deal more.

Last lesson, we bring everything from this course together into templates you can reuse without starting from scratch every time.