AI sounds just as sure when it's guessing. Where you can't check its answers yourself, set rules in advance and ask one question: verified, or assumed?
Nobody knows everything. Running a business means working in areas where you aren't the expert: tax, employment law, the software your team relies on, a supplier's contract. Each has its own rules, limits and recent changes, and no one keeps all of that in their head.
Those areas are exactly where people now turn to AI. They're also where it's hardest to tell when the answer is wrong.
Confident isn't the same as right
AI sounds the same whether it knows something or is guessing. Confident and wrong reads exactly like confident and right.
In your own field, you'd catch the mistake. Outside it, you can't, because catching it would take the knowledge you were asking for. So AI's mistakes land in the places where you're least able to spot them. That's the real pitfall: trusting it most where you can check it least.
Guard rails, set up in advance
Most businesses now use AI every day, for drafting, research, planning and a growing share of production work. Working well with it is a skill in itself, and much of that skill is knowing where you're exposed.
So don't rely on spotting errors in the moment. Set rules in advance for the areas where you know you can't. These will differ on your use case, but here are some examples:
Claims about your own business. Your figures, contracts, policies, documents. If AI tells you what's in them, it should work from the actual file, not from what it assumes a business like yours would have. Give it the source, and ask it to point to where it found the answer.
Claims about how the outside world works. A regulation, a tax rule, a supplier's terms, what a piece of software can do or what it costs. These change, and AI's knowledge has a cut-off date. Before you act, check against the official source or the provider's current documentation. If you can't check it yet, treat it as an assumption and call it one.
Instructions on how to use software. It’s never been so easy to get an answer on how something works. Where to click, which setting to change to get to where you need. When you ask AI for help with software, make sure it knows what it’s talking about. Screens change constantly. Ask it to work from a screenshot or the provider's current help pages.
When the rules apply
Checking everything would make AI useless. The rules apply when it matters: when you're about to act on the answer, when the step is hard to undo, or when the advice will reach a client or customer.
A casual explanation doesn't need checking. A change to a contract clause, your payroll settings or a customer-facing policy does.
One question that does most of the work
At any point you can ask: is that verified, or assumed?
Ask AI to answer it straight. You will work more calmly once you know where you stand.
Your blind spots are where the guard rails go
Every leader has areas they can't check for themselves. Those are where AI feels most helpful and is most risky to trust blindly.
Avoiding AI isn't the answer. Knowing your blind spots is, and making AI show its working whenever an answer lands in one.
Next time an AI gives you an answer you're about to act on, ask it: is that verified, or assumed?
